Gated Recurrent Units

Gait classification has been extensively employed in the context of cerebral palsy (CP) to facilitate clinical decision-making and to assess various treatment outcomes. However, relying solely on clinical judgment can introduce inconsistencies. This study assessed the performance of deep learning models, i.e., LSTM, GRU, and Attention-LSTM in automatic classification of the CP gait patterns, based on Rodda's classification system, in order to enhance both objectivity and accuracy. Kinematic data from 317 CP patients (634 limbs for two trials) were recorded using a Vicon motion capture system, focusing on hip, knee, and ankle angles. The data were preprocessed, normalized, and then segregated into training, validation, and test sets. The deep learning models provided moderate classification performances, in a range of 62% to 73%. The LSTM model delivered the highest accuracy (73.22%), followed by GRU (70.86%), and Attention-LSTM (62.55%). The higher performance of the LSTM model was attributed to its ability to retain long-term temporal information. It was concluded that the deep learning models are promising tools for automatic classification of the gait pattern of CP individuals and can be well employed as a decision support system to help unexperienced clinicians. However, for more accurate classification results, they need a larger dataset of the kinematics data of CP individuals and their gait trials, as well as an even sample distribution across CP groups.

The increasing volume of network traffic data exchanged among interconnected devices on the internet of things (IoT) poses a significant challenge for conventional intrusion detection systems (IDS), especially in the face of evolving and unpredictable security threats. It is crucial to develop adaptive and effective IDS for IoT to mitigate false alarms and ensure high detection accuracy, particularly with the surge in botnet attacks. These attacks have the potential to turn seemingly harmless devices into zombies, generating malicious traffic that disrupts network operations. This paper introduces a novel approach to IoT intrusion detection, leveraging machine learning techniques and the extensive UNSW-NB15 dataset. Our primary focus lies in designing, implementing, and evaluating machine learning (ML) models, including K-nearest neighbors (KNN), random forest (RF), long short-term memory (LSTM), and gated recurrent unit (GRU), against prevalent botnet attacks. The successful testing against prominent Bot-net attacks using a dedicated dataset further validates its potential for enhancing intrusion detection accuracy in dynamic and evolving IoT landscapes.

Plasmodium falciparum is a parasitic protozoan that can cause malaria, which is a deadly disease. Therefore, the accurate identification of malaria parasite mitochondrial proteins is essential for understanding their functions and identifying novel drug targets. For classifying protein sequences, several adaptive statistical techniques have been devised. Despite significant gains, prediction performance is still constrained by the lack of appropriate feature descriptors and learning strategies in current systems. Moreover, good ground truth data is important for Artificial Intelligence (AI)-based models but there is a lack of that data in the literature. Therefore, in this work, we propose a novel hybrid network that combines 1D Convolutional Neural Network (CNN) and Bidirectional Gated Recurrent Unit (BGRU) to classify the malaria parasite mitochondrial proteins. Furthermore, we curate a sequential data that are collected from National Center for Biotechnology Information (NCBI) an...

A Stock market collapse occurs when stock prices drop by more than 10% across all main indexes. Predicting a stock market crisis is difficult because of the increased volatility in the stock market. Stock price drops can be triggered by a variety of factors, including corporate results, geopolitical tensions, financial crises, and pandemic events. For scholars and investors, predicting a crisis is a difficult endeavor. We developed a model for the prediction of stock crisis using Hybridized Feature Selection (HFS) approach. Firstly, we went for the suggestion of the HFS method for the removal of stock’s unnecessary financial attributes. The Naïve Bayes approach, on the other hand, is used for the classification of strong fundamental stocks. In the third step, Stochastic Relative Strength Index (StochRSI) is employed to identify a stock price bubble. In the fourth step, we identified the stock market crisis point in stock prices through moving average statistics. The fifth is the pre...

Recently, chatbots are having a great importance in different domains and are becoming more and more common in customer service. One possible cause is the wide variety of platforms that offer the natural language understanding as a service, for which no programming skills are required. Then, the problem is related to which platform to use to develop a chatbot in the educational domain. Therefore, the main objective of this paper is to compare the main natural language understanding (NLU) engines and determine which could perform better in the educational domain. In this way, researchers can make more justified decisions about which NLU engine to use to develop an educational chatbot. Besides, in this study, six NLU platforms were compared and performance was measured with the F1 score. Training data and input messages were extracted from Mariateguino Bot, which was the chatbot of the José Carlos Mariátegui University during 2018. The results of this comparison indicates that Watson Assistant has the best performance, with an average F1 score of 0.82, which means that it is able to answer correctly in most cases. Finally, other factors can condition the choice of a natural language understanding engine, so that ultimately the choice is left to the user.

Consumers' acceptance of a food product hinges on its taste. Culinary practitioners typically conduct organoleptic tests to evaluate a food/beverage's taste. Organoleptic tests have a subjective nature, making a clear description difficult. In this study, we suggest implementing a brain signal-based electroencephalogram (EEG) taste assessment system to evaluate consumer responses to the tastes of a drink, specifically sour and sweet. The system distinguishes flavors based on EEG data. These classifiers, including recurrent neural network (RNN), long-short term memory (LSTM), and gated recurrent unit (GRU), are utilized for the classification process. Total 35 participants' EEG data were recorded for this study. Temporal (T3 and T4) and centro parietal (CP1 and CP2) channels are used for recording. EEG signal processing involves filtering, artefact elimination, and band decomposition into delta, theta, alpha, beta, and gamma frequencies. In the time domain of clean EEG data, mean absolute value, standard deviation, and variance are used for signal feature extraction. Several classifiers (RNN, LSTM, and GRU) will be fed with the signal feature values as input. An accuracy of 88.62% was achieved using LSTM in the classification. The RNN and GRU models achieved classification accuracies of 88.56% and 87.15% respectively.

În martie 2019 s-a lansat prima bază de date din industrie care urmărește lansările de dispozitive 5G la nivel mondial. Una dintre cele mai importante inovații în arhitectura 5G este virtualizarea completă a rețelei de bază (5GCN). De exemplu, „softwarizarea” funcțiilor de rețea va permite o portabilitate mai ușoară și o flexibilitate mai mare a sistemelor și serviciilor de rețea (Separarea planului de control-utilizator, CUPS). Noile caracteristici și îmbunătățiri aduse de versiunea 16 a specificației GPP sunt reflectate în actualizări și adăugiri la funcțiile Core 5G

Software development for project success. Conventional bug detection techniques like human code review and rule-based inspection are powerless to deal with complexity within large dynamic codebases. This article introduces a machine learning-based bug predictor system that entails the application of neural networks via a Gated Recurrent Unit (GRU) model to identify potential bugs in software. The GRU model is trained upon the source code, commit history, and bug reports of popular code bases such as GitHub and GitLab. The system employs token-based processing of data through the tokenization, text homogenization, and feature extraction processes. Preprocessed data is used to input into the GRU model in order to predict bugs. The performance of the model is measured with impressive metrics such as accuracy of 98.75%, precision of 97.90%, recall of 96.62%, F1-score of 97.45%. The results show that the GRU-based model performs well in code bug prediction, improving software quality, and lowering maintenance costs in the future. Some of the limitations of bug prediction using neural networks, including high computational cost, un-interpretable models, and large amounts of labeled data, are also discussed in the paper. Methods like model pruning, transfer learning, and explainable AI are debated controversially regarding overcoming limitations and being capable of being utilized more effectively in realworld applications.

Humans moving in crowded spaces adapt their pace or alter their initial path. A variety of sociocultural factors and personal preferences influence these interactions. With the development of autonomous moving platforms that need to share their physical environment with humans, this task has become increasingly valuable. In this paper, we present a way to conveniently evaluate the ability of a model to predict social interactions between pedestrians. In this manner, we conduct experiments on two prediction models: Social long short-term memory (LSTM) and Vanilla LSTM models. We use real datasets by generating synthetic datasets for which we can define the impact of social interactions on the motion of pedestrians. These hand-tailored datasets exclude individual interactions and focus on the interactions between individuals. By comparing the models’ performances on these datasets, we show that while the Social LSTM model can predict social interactions, the Vanilla LSTM model cannot. For our analysis, we introduce evaluation metrics that focus on the interactions between pedestrians, and these metrics go beyond the commonly used average and final displacement error for trajectory prediction. In particular, we analyze the prediction errors in regions of trajectories highly influenced by social interactions. Furthermore, we analyze the collision behavior of the model’s predictions and classify trajectories concerning their degree of non-linearity. This allows us to compare the models’ performances on motions differently influenced by social interactions.

Proteins perform critical functions, and their role is closely associated with their composition. Reliable prediction of protein function using computational techniques is becoming necessary because experimental limitations make it challenging to cover the massive rate of discovered proteins and their genetic transformation. This paper surveys the protein function prediction briefly discusses the employed literature's workflow, and determines if the machine and deep learning techniques have been widely used for similar objectives. Furthermore, the paper discusses the trends of protein datasets and their features and summarises researchers' questions using these datasets. Many features have been identified and extracted, ranging from traditional physicochemicals of amino acids and techniques for selecting features and reducing the dimensionality. Distinct from previous study manuscripts, we keep a detailed review of performance evaluation metrics and compare the employed protein function prediction methods, concluding the need for efficient, effective, and adaptable protein function prediction methods in real-world scenarios. We describe the machine learning processes and their development from elementary algorithms, such as logistic regression, to more sophisticated methods, like conventional and highly developed sequential deep neural networks. The techniques of computing the hyper-parameters adopted to improve the prediction efficiency have been discussed. Many studies reviewed the implemented machine and deep learning approaches for protein function prediction compared with other methods. The critical challenge well noted in protein function prediction is getting relevant information. Then, the approaches evaluated and provided future study prospects based on the results drawn from previous studies.
This review provides important information as well as prospects. We already presented this research as a preprint.

Wireless networks can be seen as the essential element of contemporary communication systems, connecting, in one way or another, billions of people and technologies all over the world. As a result, there is more of requirement from the area of application for models, which should be able to help in the analysis of the time series of mobile traffic to enhance the quality of service (QoS) in the present networks as well as in the future ones. The primary objective of this article is to develop effective artificial intelligence (AI) models for traffic load prediction in cellular networks. To achieve this, we employ three models; gated recurrent unit (GRU), bidirectional long short time memory (BiLSTM), and long short time memory (LSTM), to make numerical estimates of the network traffic at 4G long-term evolution (LTE) cell towers. The empirical results indicate that the BiLSTM model outperforms both the LSTM and GRU models, achieving root mean squared error (RMSE), mean absolute error (MAE), and R2 values of 86.64, 67.12, and 93.23%, respectively. Although this research focuses on traffic modeling for 4G LTE networks, the proposed models hold significant value for the development and optimization of the upcoming generations.

The population of elderly is growing and is projected to outnumber the youth in the future. Many researches on elderly assisted living technology were carried out. One of the focus areas is activity monitoring of the elderly. AReM dataset is a time series activity recognition dataset for seven different types of activities, which are bending 1, bending 2, cycling, lying, sitting, standing and walking. In the original paper, the author used a many-to-many Recurrent Neural Network for activity recognition. Here, we introduced a time series classification method where Gated Recurrent Units with many-to-one architecture were used for activity classification. The experimental results obtained showed an excellent accuracy of 97.14%.

This paper focuses on studying the development of an integrated versioning system for the source files used in the software project development. The server component is a .net Web service, powered by a NoSQL Type Database Management System. All functional and data resources are resident on a Cloud platform. Choosing this solution (NoSQL in conjunction with Cloud technology) provides a great boost in speed, necessary as this type of application is subject to a high density of requests, more-over so when operations like file uploads, file parsing, file content comparisons and file downloads occur at a constantly increasing rate .

This article proposes a novel hybrid network integrating three distinct architectures-CNN, GRU, and LSTMto predict stock price movements. Here Combining Feature Extraction and Sequence Learning along with Complementary Strengths can improve Predictive Performance. CNNs can effectively identify short-term dependencies and relevant features in time series, such as trends or spikes in stock prices. GRUs are designed to handle sequential data. They are particularly useful for capturing dependencies over time while being computationally less expensive than LSTMs. In the hybrid model, GRUs help maintain relevant historical information in the sequence without suffering from vanishing gradient problems, making them more efficient for long sequences. LSTMs excel at learning long-term dependencies in sequential data, thanks to their memory cell structure. By retaining information over longer periods, LSTMs in the hybrid model ensure that important trends over time are not lost, providing a deeper understanding of the time series data. The novelty of the 1D-CNN-GRU-LSTM hybrid model lies in its ability to simultaneously capture short-term patterns and long-term dependencies in time series data, offering a more nuanced and accurate prediction of stock prices. The data set comprises technical indicators, sentiment analysis, and various aspects derived from pertinent tweets. Stock price movement is categorized into three categories: Rise, Fall, and Stable. Evaluation of this model on five years of transaction data demonstrates its capability to forecast stock price movements with an accuracy of 0.93717. The improvement of proposed hybrid model for stock movement prediction over existing models is 12% for accuracy and F1-score metrics.

Stock price movement prediction is challenging due to unpredictable fluctuations and the significant impact of market sentiment and news. Accurate prediction models can enhance investor decision-making and control over stock price movements. Creating a model for predicting high-accuracy stock price movements can improve investor control over stock prices. In this study, a wide range of technical indicators and various aspects of sentiment analysis in tweets were used to predict stock price movement. The impact of the maximum number of positive comments on Tesla stocks on price increases is investigated. Also, we proposed a method for adding sentiments to each tweet. Extracted advanced sentiment analysis features such as the number of positive comments, the number of negative comments, the average score of positive comments, the average score of negative comments, daily tweet volume, ratio of positive to negative tweets. Effect of time windows with variate size is investigate. A CNN-LSTM deep neural network is used to predict stock price movement and compared with LSTM and GRU models. According to the results, the proposed CNN-LSTM deep neural network has the best results to predict stock price movement over the 30-day interval.
INDEX TERMS: Long short-term memory (LSTM), convolutional neural network (CNN), gated recurrent unit (GRU), stock price movement prediction, sentiment analysis, Twitter.

Climate change, increasing food demand for a growing population, and industrial expansion has placed tremendous stress on crop yield. An accurate crop yield prediction model is vital for making informed decisions regarding market prices and ensuring sustainable agricultural practices. Existing schemes for crop yield prediction and recommendation suffer from insufficient accuracy and high computational time due to suboptimal parameter settings. To address these challenges, this study introduces an effective decision support system that integrates a high-level feature learning-based ensemble of fine-tuned deeplearning models. The system focuses on crop yield prediction and profitable crop recommendation. A one-dimensional depth-wise separable convolution autoencoder (1D-DSCAE) is employed for feature extraction, and a Coati-optimized gated recurrent unit (CO-GRU) is used for crop yield prediction. Additionally, a Fuzzy Inference System (FIS) is integrated for decision-making to recommend profitable crops based on predicted yield and market prices. The proposed model demonstrates superior performance in terms of accuracy, precision, recall, F1 score, error, loss, and computational efficiency.

Rezumat / Abstract
Rezumat în limba română: Lucrarea analizează securitatea unor site-uri standard construite pe platforma WordPress și compară evoluția platformei între două versiuni majore (5.0 și 5.4). O primă parte principală se referă la testarea la nivel de server (serverul https://olivian.ro). Urmează o parte în care se face o testare la nivel de instanță WordPress, se testează cu diferite soluții de securitate două instalări de WordPress, WordPress 5.0, https://olivian.ro/wp50/, și WordPress 5.4, https://olivian.ro/wp54/. Lucrarea se încheie cu soluții pentru o parte din problemele observate.
Domeniu principal: Securitate aplicații web.
Domenii adiacente: Programare, găzduire, managementul aplicațiilor web.
Tipul lucrării: cercetare.

English language abstract: The paper analyzes the security of standard-built web sites on the WordPress platform and compares the evolution of the platform between two major versions (5.0 and 5.4). The first main part refers to testing at the server level (https://olivian.ro server). Next, a WordPress instance-level test is performed: two WordPress installations are tested with different security solutions - WordPress 5.0, https://olivian.ro/wp50/, and WordPress 5.4, https://olivian.ro/wp54/. The paper finishes with solutions to some of the noticed problems.
Main field: Web application security.
Other fields: Programming, hosting, web application management
Type of work: research.
 
Introducere
În lucrarea de față am avut ca obiectiv o comparație între securitatea platformei de gestionare a conținutului WordPress pentru două versiuni de WordPress distincte - 5.0 și 5.4. Am urmărit în paralel să văd cum a evoluat securitatea versiunii de WordPress între două versiuni destul de importante, și, la data începerii cercetării (aprilie 2020), recente - WordPress 5.0 a fost lansat în 6 decembrie 2018, iar în 31 martie 2020 a apărut WordPress 5.4. Am testat inițial serverul de pe hosting-ul site-ului propriu, olivian.ro (testare la nivel de server), ulterior am testat două instalări de WordPress distincte - WordPress 5.0, aici: https://olivian.ro/wp50/, și WordPress 5.4, aici: https://olivian.ro/wp54/ (testare la nivel de instanță). La final, am căutat soluții pentru problemele găsite.
De ce am ales tema - securitate pe WordPress? Experiența personală cu WordPress a început cu un site pe WordPress.com (e diferit față de WordPress.org) al cărui prim articol a fost în 19 noiembrie 2007 și, ulterior, a ajuns la vreo 40 de articole. Am continuat cu un blog pe domeniul propriu, bazat pe WordPress-ul pe care îl vom analiza în lucrarea de față, WordPress.org, prin martie 2008. Am început, pe măsură ce foloseam WordPress.org, să folosim platforma nu atât pentru a crea bloguri, cât site-uri de prezentare pentru diferite entități (ONG-uri, școli, clienți în colaborare cu un PFA - Persoană Fizică Autorizată). Am înregistrat ulterior mai multe domenii web, împărțite între tipuri de activități și diferite de limbi în care erau scrise articolele, urmând ca în 9 aprilie 2012 să înregistrăm olivian.ro și să unim toate vechile platforme pe una singură, pe care lucrăm și astăzi.
În toamna anului 2018 am început o colaborare ca dezvoltator pe platforma WordPress pentru o agenție de web design și development din București, activitate pe care o desfășurăm și la data scrierii acestei lucrări (luna mai 2020). Am lucrat mult mai intens, astfel, pe platforma WordPress.
Am lucrat la lansarea a mai mult de 100 de site-uri pe platforma WordPress,  cu intensitate a implicării diferită (instalarea unei teme/template pentru un site deja lansat, reinstalarea unei alte teme pentru un site deja lansat, site realizat de la zero - scris cod, contribuție semnificativă la realizare site). De asemenea, am gestionat de-a lungul timpului (mentenanță WordPress - actualizări site-uri) câteva zeci de site-uri. De asemenea, am folosit din perspectiva de blogger, autor de articole, platforma WordPress pentru a scrie câteva mii de articole pe diferite platforme, avem așadar și experiență pe partea de folosire a platformei.
Pe partea de programare / cunoștințe Internet, în afara programului de masterat curent (Universitatea "Titu Maiorescu" din București - Masterat Securitatea sistemelor informatice și a rețelelor informaționale), am absolvit un liceu profil informatic, am urmat cursurile unui masterat în IT (Modelare și Tehnologii Informatice - Universitatea „Ovidius” din Constanța), și am lucrat aproape continuu din martie 2006 la diferite locuri de muncă care presupuneau folosirea intensă a Internetului (în principal - SEO, Search Engine Optimization - optimizare pentru motoarele de căutare, dar și uzabilitate, UX - User Experience).
Ca notă de redactare a lucrării, pentru un paragraf în care am citat exact, sau aproximativ, o resursă, am pus la final de paragraf sursa. Dar dacă după un paragraf urma o enumerare, am pus sursa doar la paragraful de sus, chiar la finalul lui, lăsând să se înțeleagă că enumerarea face parte din paragraful citat și conține referința. Am evitat astfel să punem sursa la fiecare item al unei enumerări, nu ar fi fost plăcut vizual ca dacă e o enumerare cu 5 elemente cu aceeași sursă, să menționăm sursa la fiecare element din cele 5.

Parkinson’s Disease (PD) is a progressive neurodegenerative disorder that causes loss of motor functions, emphasizing the need for prompt and accurate detection. This study aimed to design an automated system for early PD detection focusing on vocal features, as verbal disturbances often appear before more noticeable physical symptoms. A multisource dataset comprising speech recordings, biomedical voice metrics, and replicated voice recordings was utilized for this task. Several data augmentation techniques were explored to address data imbalances and size limitations, with Grouped Synthetic Minority Over-Sampling Technique (SMOTE) and Conditional Generative Adversarial Networks (CGAN) with Structured Conditioning were identified as the most effective. Among validated feature reduction methods, Boruta emerged as the optimal approach. This study focused on hybrid models that integrated Deep Learning (DL) and Machine Learning (ML), including attention mechanisms. Performance analysis identified the Attention-based Long Short-Term Memory - Bidirectional Gated Recurrent Unit (LSTM-BiGRU) as the superior model, attaining accuracies of 97.50%, 96.92%, and 96.25% across the datasets, respectively. Validation with K-fold and Leave-One-Participant-Out (LOPO) Cross-Validation (CV) ensured the model’s generalizability. Additionally, Explainable Boosting Machines (EBM) were employed to interpret the model’s decision-making process, enabling accurate and interpretable diagnostic tools for early PD detection. Keywords—Parkinson’s Disease, Speech and Vocal Features, LOPO CV, Boruta, Attention LSTM-BiGRU, EBM.

In recent years, there has been a significant increase in the utilization of renewable resources for electricity generation. Consequently, accurate short-term forecasting of renewable power production has become crucial for power system operations. However, Renewable Power Production Forecasting (RPPF) presents unique challenges due to the intermittent and uncertain nature of renewable energy sources. This paper proposes a novel approach to short-term RPPF. The proposed model integrates various techniques, including Long Short-Term Memory (LSTM), Gated Recurrent Unit (GRU), Autoregressive Integrated Moving Average (ARIMA), Multi-Layer Perceptron (MLP), and Adaptive Neuro-Fuzzy Inference System (ANFIS). The aim is to enhance the accuracy and predictive performance of renewable power production forecasts. The suggested hybrid model employs the Modified Relief-Mutual Information (MRMI) feature selection technique to identify the most influential input data for prediction. Subsequently, the combined model generates a 24-hour ahead RPP prediction using a weighted output approach. By capitalizing on the strengths of each individual model, the combined method mitigates their weaknesses, thereby improving the overall efficiency of the forecasting process. The accuracy and performance of the proposed method are evaluated through two case studies involving solar farm power generation at the Mahan, Iran and Rafsanjan, Iran sites. The results demonstrate the effectiveness of the hybrid model in enhancing the accuracy of short-term RPPF. By combining multiple forecasting methods and utilizing the MRMI feature selection technique, the proposed method significantly improves prediction accuracy.

These improvements open many possibilities in solving Natural Language Processing downstream tasks. Such tasks include machine translation, speech recognition, information retrieval, sentiment analysis, summarization, question answering, multilingual dialogue systems development, and many more. Language models are one of the most important components in solving each of the mentioned tasks. This paper is devoted to research and analysis of the most adopted techniques and designs for building and training language models that show a state of the art results. Techniques and components applied in the creation of language models and its parts are observed in this paper, paying attention to neural networks, embedding mechanisms, bidirectionality, encoder and decoder architecture, attention, and self-attention, as well as parallelization through using transformer. As a result, the most promising techniques imply pre-training and fine-tuning of a language model, attention-based neural netwo...

Plasmodium falciparum is a parasitic protozoan that can cause malaria, which is a deadly disease. Therefore, the accurate identification of malaria parasite mitochondrial proteins is essential for understanding their functions and identifying novel drug targets. For classifying protein sequences, several adaptive statistical techniques have been devised. Despite significant gains, prediction performance is still constrained by the lack of appropriate feature descriptors and learning strategies in current systems. Moreover, good ground truth data is important for Artificial Intelligence (AI)-based models but there is a lack of that data in the literature. Therefore, in this work, we propose a novel hybrid network that combines 1D Convolutional Neural Network (CNN) and Bidirectional Gated Recurrent Unit (BGRU) to classify the malaria parasite mitochondrial proteins. Furthermore, we curate a sequential data that are collected from National Center for Biotechnology Information (NCBI) and UniProtKB/ Swiss-Prot proteins databanks to prepare a dataset that can be used by the research community for AI-based algorithms evaluation. We obtain 4204 cases after preprocessing of the collected data and denote this set of proteins as PF4204. Finally, we conduct an ablation study on several conventional and deep models using PF4204 and the benchmark PF2095 datasets. The proposed model 'CNN-BGRU' obtains the accuracy values of 0.9096 and 0.9857 on PF4204 and PF2095 datasets, respectively. In addition, the CNN-BGRU is compared with state-of-the-arts, where the results illustrate that it can extract robust features and identify proteins accurately.

Appointment booking is a part of everyday life nowadays. From booking movie show tickets to booking flight seats everything is online. In recent times the pandemic has brought some tough challenges before the healthcare systems which include appointment booking. As hospitals and clinics witness an overwhelming surge of patients the healthcare workers are found to be overburdened. Oftentimes important calls made by patients to the hospitals go unanswered or result in long wait times. Speech is the primary mode of communication among human beings. Many patients would prefer to call-in and book an appointment before visiting the crowded hospitals. This paper is focused on developing an AI voice bot which works on call to communicate with patients and subsequently book appointments and/or answer questions related to that hospital. Automatic Speech Recognition (ASR) is the process of deriving the transcription of an utterance, given in the speech waveform. This paper aims at completely a...

Electronic Health Records (EHR) data analysis plays a crucial role in healthcare system quality. Because of its highly complex underlying causality and limited observable nature, causal inference on EHR is quite challenging. Deep Learning (DL) achieved great success among the advanced machine learning methodologies. Nevertheless, it is still obstructed by the inappropriately assumed causal conditions. This work proposed a novel method to quantify clinically well-defined causal effects as a generalized estimation vector that is simply utilizable for causal models. We incorporated it into DL models to achieve better predictive performance and result interpretation. Furthermore, we also proved the existence of causal information blink spots that regular DL models cannot reach.

Food image recognition is essential for developing an elderly-friendly dailyfood recording application in Indonesia. However, existing datasets andmodels are limited and do not cover the diversity and complexity ofIndonesian food. In this paper, we present a new dataset of 24,427 images of160 types of Indonesian food with higher variety and quality than previousdatasets. We also train and compare the performance of 67 models based on16 state-of-the-art deep learning architectures on this dataset. We find thatefficientnet_v2_l provides the best accuracy of 85.44%, followed by othermodels such as convnext_large and swin_s. We also discuss the trade-offbetween model size and performance, as well as the challenges andlimitations of food image classification. Our dataset and models can serve asa basis for developing a user-friendly and accurate food recordingapplication for the elderly population in Indonesia.

Forecasting stock prices, a task complicated by the inherent volatility of the stock market, poses a significant challenge. The ability to accurately forecast stock prices is crucial, as it provides investors with crucial insights, enabling them to make informed strategic decisions. In this paper, we propose a novel investment strategy that relies on predicting stock prices. Our approach utilizes a hybrid predictive model that combines light gradient-boosting machine (LightGBM) and extreme gradient boosting (XGBoost). This model is designed to generate short to medium-term forecasts for a wide range of stocks. The strategy has shown promising results, surpassing the local market indices used as benchmarks in terms of both risk and return. Our findings demonstrate the strategy's effectiveness in both upward and downward market trends, underscoring its potential as a robust tool for portfolio management in diverse market conditions.

The internet of things (IoT) greatly impacts daily life by enabling efficient data exchange between objects and servers. However, cyber-attacks pose a serious threat to IoT devices. Intrusion detection systems (IDS) are vital for safeguarding networks, and machine learning methods are increasingly used to enhance security. Continuous improvement in accuracy and performance is crucial for effective IoT security. Deep learning not only outshines traditional machine learning methods but also holds untapped potential in fortifying IDS systems. This paper introduces an innovative deep learning framework tailored for anomaly detection within IoT networks, leveraging bidirectional long short-term memory (BiLSTM) and gated recurrent unit (GRU) architectures. The hyper parameters of the proposed model are optimized using the JAYA optimization technique. These models are validated using IoT-23 and MQTTset datasets. Several performance metrics including accuracy, precision, recall, F-score, true negative rate (TNR), false positive rate (FPR), and false negative rate (FNR), have been selected to assess the effectiveness of the suggested model. The empirical results are scrutinized and juxtaposed with prevailing approaches in the realm of intrusion detection for IoT. Notably, the proposed method emerges as showcasing superior accuracy when contrasted with existing methods.

Floods are a major global problem affect communities and businesses. For these effects to be mitigated and emergency measures to be improved, accurate prediction is essential. Conventional flood prediction models frequently fail because the models ignore important hydrological elements like water discharge and instead solely use rainfall data. This limitation was
addressed by the combination of rainfall and water discharge data on internet
of things (IoT)-based technologies. It focuses on analyzing historical records from flood-prone areas in Semarang using gated recurrent unit (GRU) models. The findings demonstrate how effectively the GRU model performs when rainfall and water discharge factors are taken into account, resulting in very accurate and dependable predictions of flood events. Precision, Recall, and F1-score are evaluation metrics that demonstrate the accuracy on which the model determines flood emergency statuses. This study advances flood prediction methods and highlights the value of integrating internet of things data to improve preparedness and resilience against flood disasters.

The article focuses on enhancing stock market price prediction through artificial neural networks and machine learning. It underscores the significance of improving forecast accuracy by incorporating historical stock prices, macroeconomic indicators, news events, and technical indicators. Exploring deep learning principles, it delves into convolutional neural networks (CNN), recurrent neural networks (RNN), including long shortterm memory (LSTM), and gated recurrent unit (GRU) modifications. This financial time series processing study covers data preprocessing, creating training/test sets, and selecting evaluation metrics. Results suggest promising applications for the developed forecasting models in stock markets, stressing the importance of considering various factors for precise forecasts in dynamic financial environments. Historical reserve data serves as the model foundation. Integration of macroeconomic, news, and technical indicators offers a holistic approach, aiding trend and anomaly identification for enhanced forecasts. The article recommends suitable deep learning architectures, highlighting LSTM and GRU's effectiveness in adapting to intricate data dependencies. Experimental outcomes showcase these architectures' benefits in predicting stock market prices, offering valuable insights for finance and asset management professionals in financial analysis and machine learning realms.

Dyslexia in children are serious problems that need to be addressed early. Many previous studies have focused on the detection/prediction of dyslexia. However, in the prediction process, there is often an imbalance in the dataset used (between patients with dyslexia and non-dyslexia). Therefore, we are trying to build a system using recurrent neural networks architectures that can quickly and accurately predict the possibility of a child having dyslexia. To overcome the data imbalance between dyslexics and non-dyslexics, we also apply the synthetic minority oversampling technique (SMOTE) method to the dataset. SMOTE will synthesize dyslexic data to balance the numbers with non-dyslexic data. This study used a dataset of 3640 participants (392 dyslexic and 3248 non-dyslexics). For the process of predicting dyslexia, several algorithms such as simple recurrent neural networks (RNN), long short termmemory (LSTM), and gate recurrent units (GRU) are used. As a result, there is an increase in prediction accuracy when SMOTE is applied (compared to without SMOTE) in the dyslexia forecasting process using RNN (92.68% for training and 91.16% for testing), LSTM (94.81% for training and 93.16% for testing), and GRU (96.43% for training and 92.24% for testing). Using SMOTE+RNN architecture in this research increased the accuracy of dyslexia prediction by up to 5% compared to without SMOTE.

Accident detection is a vital part of traffic safety. Many road users suffer from traffic accidents, as well as their consequences such as delay, congestion, air pollution, and so on. In this study, we utilize two advanced deep learning techniques, Long Short-Term Memory (LSTM) and Gated Recurrent Units (GRUs), to detect traffic accidents in Chicago. These two techniques are selected because they are known to perform well with sequential data (i.e., time series). The full dataset consists of 241 accident and 6,038 non-accident cases selected from Chicago expressway, and it includes traffic spatiotemporal data, weather condition data, and congestion status data. Moreover, because the dataset is imbalanced (i.e., the dataset contains many more non-accident cases than accident cases), Synthetic Minority Over-sampling Technique (SMOTE) is employed. Overall, the two models perform significantly well, both with an Area Under Curve (AUC) of 0.85. Nonetheless, the GRU model is observed to p...

Spam detection in SMS communication is a crucial task for maintaining the quality of messaging services and protecting users from unwanted and potentially harmful messages. Arabic SMS spam detection poses unique challenges due to the rich morphology and complex structure of the Arabic language, which can significantly impact the performance of traditional text classification methods. To address these challenges, this paper presents a novel approach for Arabic SMS spam detection using a hybrid deep learning model that combines Convolutional Neural Networks (CNN) and Bidirectional Long Short-Term Memory (Bi-LSTM) networks. The proposed model leverages the strengths of CNNs in capturing local features and patterns in the text and the capability of Bi-LSTM networks to understand long-term dependencies and contextual information. This hybrid architecture is designed to effectively handle the complexities of the Arabic language and improve the accuracy of spam detection. The model was evaluated on a dataset of Arabic SMS messages, consisting of both ham (non-spam) and spam messages. The dataset underwent preprocessing steps, including text cleaning, tokenization, and padding, to prepare it for training the deep learning model. The hybrid model was trained using the Adam optimizer and evaluated using accuracy, precision, recall, and F1 score metrics. Early stopping was implemented to prevent overfitting during the training process. The results demonstrate that the hybrid model achieved high performance, with an accuracy of 0.9699, precision of 0.9739, recall of 0.9675, and an F1 score of 0.9707. These metrics indicate the model's effectiveness in accurately detecting spam in Arabic SMS messages. Additionally, the paper provides visualizations of the confusion matrix, ROC curve, and training-validation loss graph to illustrate the model's performance. The implications of this research are significant for the field of Arabic text classification and spam detection. The proposed hybrid model offers a robust solution for accurately classifying Arabic SMS messages, which can be integrated into messaging platforms to enhance spam detection capabilities and improve user experience. Future work could explore data augmentation techniques, transfer learning, and advanced hybrid architectures to further enhance the model's performance and applicability.

It delivers the most accurate and noiseless experience utilizing the best in language and machine learning to comprehend and respond to the inquiries of people. The technical architecture of the Bing GPT Voice Assistant is covered in this paper, along with the data processing pipelines, voice recognition technology integration, and underlying machine learning models. It also examines the difficulties encountered in the development process, including protecting user privacy, responding to a variety of user inquiries, and preserving conversational context over lengthy exchanges. The work also aims to study and comprehend the design procedure, effectiveness and efficiency, which aids in the progression of audio technology. Voice technology users get the advantage of voice assistance because the work helps in the process's continuation by enabling readers to know the design and working.

Android is increasingly being targeted by malware since it has become the most popular mobile operating system worldwide. Evasive malware families, such as Chamois, designed to turn Android devices into bots that form part of a larger botnet are becoming prevalent. This calls for more effective methods for detection of Android botnets. Recently, deep learning has gained attention as a machine learning based approach to enhance Android botnet detection. However, studies that extensively investigate the efficacy of various deep learning models for Android botnet detection are currently lacking. Hence, in this paper we present a comparative study of deep learning techniques for Android botnet detection using 6802 Android applications consisting of 1929 botnet applications from the ISCX botnet dataset. We evaluate the performance of several deep learning techniques including: CNN, DNN, LSTM, GRU, CNN-LSTM, and CNN-GRU models using 342 static features derived from the applications. In ou...

One indicator of a university's educational quality is the proportion of enrolled students who actually graduate within four years. This proportion is typically fewer than the number of students that enroll in a given year. A low graduation rate can have a negative impact on both the university's reputation and its accreditation because it indicates that fewer students are completing their degrees. Student activity, economic, and other issues all play a role in why some students are unable to complete their degrees on time. As a result, stakeholders need a model that can predict whether or not students will graduate on time as a means of evaluating and giving a basis for policy actions. This research proposes a model for converting textual data into an image format using a deep learning convolutional neural network (CNN), and then classifying the extracted features using a variety of machine learning classification algorithms like the decision tree, random forest, Naive Bayes, support vector machine (SVM), and k-nearest neighbor (K-NN). The classification model trained on feature extraction data had a 96.1% accuracy rate, while the classification model trained on the original data achieved a 71.2% accuracy rate.

Predicting highrisk vascular diseases is a significant issue in the medical domain. Most predicting methods predict the prognosis of patients from pathological and radiological measurements, which are expensive and require much time to be analyzed. Here we propose deep attention models that predict the onset of the high risky vascular disease from symbolic medical histories sequence of hypertension patients such as ICD-10 and pharmacy codes only, Medical History-based Prediction using Attention Network (MeHPAN). We demonstrate two types of attention models based on 1) bidirectional gated recurrent unit (R-MeHPAN) and 2) 1D convolutional multilayer model (C-MeHPAN). Two MeHPAN models are evaluated on approximately 50,000 hypertension patients with respect to precision, recall, f1-measure and area under the curve (AUC). Experimental results show that our MeHPAN methods outperform standard classification models. Comparing two MeHPANs, R-MeHPAN provides more better discriminative capabi...

In this era of technology, a communication barrier is a thing of the past. With each passing day, different types of language-based applications are being launched. There are 109 official languages Google has translated to date. However, the Dzongkha translation has not been studied. The purpose of this paper was to study Dzongkha to English translation. The parallel corpus was collected from the Dzongkha development commission of Bhutan. The dataset consisted of 53018 sentence pairs. Unique words in Dzongkha and English were 13,393 and 12,506 respectively. Different neural machine translation models were implemented. The experimental results show that the bleu score of Seq2Seq models followed a fluctuating trend. However, the bleu score of the transformer model increases gradually. It was observed that the transformer outperformed the Seq2Seq models. The highest accuracy and the lowest training loss obtained were 84.46% and 0.014858 respectively with a bleu score of 64.89.

Inspired by the adequacy of convolutional neural networks in implicit extraction of visual features and the efficiency of Long Short-Term Memory Recurrent Neural Networks in dealing with long-range temporal dependencies, we propose a Convolutional Long Short-Term Memory Recurrent Neural Network (CNNLSTM) for the problem of dynamic gesture recognition. The model is able to successfully learn gestures varying in duration and complexity and proves to be a significant base for further development. Finally, the new gesture command TsironiGR-dataset for human-robot interaction is presented for the evaluation of CNNLSTM.

Text summarization involves creating a concise version of a text while preserving its essential details and core message. This technique allows for quick understanding of lengthy articles, documents, or books, saving time and enhancing comprehension. There are two primary approaches to summarization: extractive and abstractive. Extractive summarization involves selecting key sentences or phrases from the text, while abstractive summarization generates new sentences that convey the original meaning. In the context of Urdu news articles, summarization is particularly valuable, as it enables readers with limited time or attention to grasp the main points quickly. This study explores various methods for summarizing Urdu news articles, evaluating both extractive and abstractive approaches using various datasets. To evaluate the proposed model, we use ROUGE metric which shows the significant improvement and efficiency compared with existing models. The study also highlights challenges and future directions in this field, including the complexity of Urdu sentence structures, addressing biases in content of news, and incorporating the latest development natural language processing and deep learning.

Nowadays, large quantities of data are being collected for analytical and forecasting purposes. They not necessarily have the same data generating process; thus, it might be difficult identifying a single forecasting model to apply to all possible scenarios. Considering a time series as an independent dataset, choosing whether to apply whether a linear or a nonlinear forecasting model separately, may lead to inaccurate results, which means inappropriate future interventions and decisions. This paper regards a simulation study for the quantitative evaluation of the dependence of the forecasting models' performance on the characteristics of the data. We have simulated several datasets with known time series characteristics, under controlled conditions, including linear data generating processes, such as Autoregressive (AR), Seasonal Autoregressive (SAR), and nonlinear ones such as Self-Exciting Threshold Autoregressive (SETAR) and Logistic Map. Heterogeneous time series are obtained by combining several data generating processes into one time series, to identify the more appropriate forecasting model for real world datasets, while the length and the number of the series are varied in different scenarios, to explore the effect of data availability on the performance of the forecasting models. We have performed forecasts using both linear and nonlinear forecasting models, an Autoregressive Integrated Moving Average (ARIMA) model, a Linear Exponential Smoothing model, an Exponential smoothing state space model with Box-Cox transformation, ARMA errors, Trend and Seasonal components (BATS) and Trigonometric seasonality Box-Cox transformation ARMA errors Trend and Seasonal components (TBATS) model and a Nonlinear Autoregressive Neural Network (NARNN) and compared their performance in terms of percentage errors.

In this paper, we propose a novel deep learning approach for predicting stock prices that combines the power of time series analysis and Monte Carlo simulation. Our approach utilizes historical stock data to train a deep learning model that can predict future stock prices with high accuracy. We use Python as our programming language and apply various deep learning techniques such as Long Short-Term Memory (LSTM) networks, Gated Recurrent Units (GRU), and Convolutional Neural Networks (CNN) to extract meaningful patterns from the time series data. To account for the uncertainty and volatility of the stock market, we incorporate Monte Carlo simulation into our approach. By generating multiple scenarios of future stock prices, we can estimate the probability distribution of future stock prices and make more informed investment decisions. We demonstrate the effectiveness of our approach by predicting stock prices for a variety of large and small capitalizations and in different markets, using both daily and up-to-the-minute frequency data. Our results show that deep learning and Monte Carlo simulation can be a powerful tool for predicting stock prices and managing investment risk. By combining these techniques, we can build a robust and efficient system that can adapt to changing market conditions and provide accurate and timely stock price predictions.

Nowadays heart diseases and their diagnosis have emerged as a prominent subject in health care systems, given that the heart performs a crucial role in the human body. Several computational techniques have been explored for the recognition and classification of cardiac diseases using Electrocardiogram (ECG) signals. Deep Learning (DL) is a present focus in healthcare solicitations, particularly in the classification of heartbeats in ECG signals. Many studies have utilized dissimilar DL models, including RNN (Recurrent Neural Networks), GRU (Gated Recurrent Unit), and CNN (Convolutional Neural Networks), to classify heartbeats using the MIT-BIH arrhythmia dataset. This article presents a methodical exploration of Bi-LSTM (Bi directional Long Short-Term Memory) based DL models for heartbeat classification using various quality metrics. Proposed variants include the Bi-LSTM model, demonstrating remarkable accuracy in classifying the heartbeats into five classes: Normal (N) beat, Supraventricular (S) beat, Ventricular contraction (V), Fusion beats (F), and Unclassifiable Beat (Q). The proposed technique outperforms present classifiers with Accuracy, Sensitivity, Specificity, and F1 score values of 98%, 96.9%, 97.4%, and 97.5% respectively. The simulations are conducted using MATLAB 2020a.

In the ever-evolving landscape of cybersecurity, the need for robust intrusion detection systems has become paramount. This paper introduces a cutting-edge intrusion detection algorithm designed to enhance network security through the integration of advanced machine learning and deep learning methodologies. The proposed algorithm capitalizes on the strengths of both paradigms to achieve a comprehensive and adaptive approach to identifying malicious activities within a network. This research focuses on enhancing network security through the development and evaluation of a novel intrusion detection system leveraging both deep learning and traditional machine learning approaches. Utilizing the NSL-KDD dataset, we employ the Long Short-Term Memory (LSTM) model, a superior version of Recurrent Neural Networks (RNNs), and the K-Nearest Neighbors (KNN) algorithm for binary and multi-class classification of network intrusion anomalies. The LSTM model excels in capturing temporal dependencies, enabling the detection of nuanced sequential patterns, while the KNN algorithm contributes to a comprehensive classification framework. Experimental results demonstrate the effectiveness of the hybrid methodology, showcasing improved accuracy, precision, and recall compared to traditional methods. This research underscores the potential of integrating deep learning and classical machine learning techniques to bolster the capabilities of intrusion detection systems in safeguarding against evolving cyber threats.

An accurate prediction of ionospheric total electron content (TEC) at the primary stage is essential for applications related to global navigation satellite systems (GNSS) under varying weather conditions. The previous TEC prediction schemes contribute for each time step that increases the prediction time. The eye contact phenomenon establishes a metaphorical connection which intends to capture and emphasize the attention worthy elements in a sequence. This research introduces a deep learning approach which is a combination of attention-based bidirectional long short-term memory and gated recurrent unit (Bi-LSTM GRU) to predict TEC in the ionosphere. Bidirectional LSTM is the better option for achieving durability when combined with a gated recurrent unit (GRU) to predict TEC in the ionosphere. The proposed approach is evaluated with the existing LSTM approach for root mean square error (RMSE) during training and validation. The RMSE while predicting the global ionospheric delay using the existing LSTM for 20 epochs is seen to be 0.004, whereas the existing approach achieves a training error of 0.003.

These improvements open many possibilities in solving Natural Language Processing downstream tasks. Such tasks include machine translation, speech recognition, information retrieval, sentiment analysis, summarization, question answering, multilingual dialogue systems development, and many more. Language models are one of the most important components in solving each of the mentioned tasks. This paper is devoted to research and analysis of the most adopted techniques and designs for building and training language models that show a state of the art results. Techniques and components applied in the creation of language models and its parts are observed in this paper, paying attention to neural networks, embedding mechanisms, bidirectionality, encoder and decoder architecture, attention, and self-attention, as well as parallelization through using transformer. As a result, the most promising techniques imply pre-training and fine-tuning of a language model, attention-based neural netwo...

The dynamic structural load identification capabilities of the gated recurrent unit, long short-term memory, and convolutional neural networks are examined herein. The examination is on realistic small dataset training conditions and on a comparative view to the physics-based residual Kalman filter (RKF). The dynamic load identification suffers from the uncertainty related to obtaining poor predictions when in civil engineering applications only a low number of tests are performed or are available, or when the structural model is unidentifiable. In considering the methods, first, a simulated structure is investigated under a shaker excitation at the top floor. Second, a building in California is investigated under seismic base excitation, which results in loading for all degrees of freedom. Finally, the International Association for Structural Control-American Society of Civil Engineers (IASC-ASCE) structural health monitoring benchmark problem is examined for impact and instant loading conditions. Importantly, the methods are shown to outperform each other on different loading scenarios, while the RKF is shown to outperform the networks in physically parametrized identifiable cases.

Social media is critical in today's world for exchanging information and disseminating ideas. A person's emotional impact has a significant impact on their day-today life. Sentiment analysis is a form of text mining that locates and pulls out subjective information from sources, allowing a company to track discussions online and monitor social sentiment about their brand, product, or service. Simply put, sentiment analysis helps determine the author's attitude towards a topic. Positive, neutral, or negative pieces of writing are classified by sentiment analysis software. Deep learning algorithms and various functions of natural language processing helps to interpret the written or spoken sentiments regarding a topic. An ecosystem where millions of bytes of data are produced daily has enabled sentiment analysis to be a key tool for interpreting these huge chunks of data. The purpose of this work is to conduct a sentiment analysis on "tweets" by making use of a variety of machine learning algorithms. The study will make an attempt to categorise the polarity of the tweet as either positive, negative, or neutral. In the event that a tweet has only positive, negative, or neutral components, the label assigned to the tweet will be determined by the sentiment that predominates.

These improvements open many possibilities in solving Natural Language Processing downstream tasks. Such tasks include machine translation, speech recognition, information retrieval, sentiment analysis, summarization, question answering, multilingual dialogue systems development, and many more. Language models are one of the most important components in solving each of the mentioned tasks. This paper is devoted to research and analysis of the most adopted techniques and designs for building and training language models that show a state of the art results. Techniques and components applied in the creation of language models and its parts are observed in this paper, paying attention to neural networks, embedding mechanisms, bidirectionality, encoder and decoder architecture, attention, and self-attention, as well as parallelization through using transformer. As a result, the most promising techniques imply pre-training and fine-tuning of a language model, attention-based neural netwo...

For over a decade, online job portals have been providing their services to both job seekers and employers in search of hiring opportunities. Because of the high demand for recruitment, it is insufficient to use conventional hiring methods to find a suitable candidate to fill the position. Validating resumes online is challenging due to the potential for manual errors, making the process inherently risky. The bidirectional method comprises named entity recognition (NER) for extracting the required resumes for recruiters. Cosine similarity shows the match percentage of resumes for the job requirements and vice versa. In an attempt to tackle an issue of unregistered words, a solution called decoder attention with pointer network (DA-PN) has been introduced. This method incorporates the use of coverage mechanism to prevent word repetition through generated text summary. DA-PN+Cover method with mixed learning objective (MLO) (DA-PN+Cover+MLO) is utilized for protecting grow of increasing faults in generated text summary. Performance of proposed method is estimated using evaluation indicator recall oriented understudy for gisting evaluation (ROUGE) and attains an average of 27.47 which is comparatively higher than existing methods.