Academia.eduAcademia.edu
All Topics
Computer Science
Artificial Intelligence
Email Spam Filter

Email Spam Filter

description105 papers
group91 followers
lightbulbAbout this topic
An email spam filter is a software application designed to identify and block unsolicited or unwanted email messages, commonly known as spam. It employs various algorithms and techniques, such as keyword analysis and machine learning, to assess the likelihood of an email being spam and to protect users' inboxes from irrelevant or harmful content.
lightbulbAbout this topic
An email spam filter is a software application designed to identify and block unsolicited or unwanted email messages, commonly known as spam. It employs various algorithms and techniques, such as keyword analysis and machine learning, to assess the likelihood of an email being spam and to protect users' inboxes from irrelevant or harmful content.

Key research themes

1. How can machine learning algorithms and feature engineering optimally detect and classify email spam?

This research area focuses on leveraging diverse machine learning models, including traditional classifiers like Naïve Bayes, SVM, Random Forest, and ensemble techniques, to improve spam email detection accuracy. A strong emphasis is placed on feature extraction and data preprocessing methods such as TF-IDF vectorization, word embeddings, and keyword analysis to enhance the discriminatory power of models. This theme matters because accurate spam detection reduces resource wastage, protects user privacy, and mitigates financial and phishing risks associated with spam emails.

Spam-Detection with Comparative Analysis and Spamming Words Extractions
by chetna kaushal

2024

Key finding: This study applied four machine learning and two deep learning models on combined datasets including TREC07 and Enron to classify spam emails and identify recurrent spam keywords. It found that advanced feature engineering... Read more
articleView PaperdownloadDownload
Email Spam Detection Using Machine Learning
by IRJET Journal

2023, IRJET

Key finding: Utilizing TF-IDF text representation combined with machine learning algorithms such as Support Vector Machines (SVM), Random Forest, and Naïve Bayes, this work demonstrated how numerical features derived from NLP techniques... Read more
articleView PaperdownloadDownload
Evaluation of Supervised Learning Models for Automatic Spam Email Detection
by Tsehay Assegie

2024, Research Square (Research Square)

Key finding: Through an empirical comparison of eight supervised models on a pre-processed and balanced email dataset, the study found Random Forest to consistently outperform others with an accuracy of 96.6%. The evaluation incorporated... Read more
articleView PaperdownloadDownload
Spam based Email Identification and Detection using Machine Learning Techniques
by Joyece Jane

2023

Key finding: This paper systematically applies and compares machine learning algorithms such as Naïve Bayes, SVM, and ensemble methods alongside bio-inspired algorithms on multiple datasets with extensive preprocessing. It confirms that... Read more
articleView PaperdownloadDownload
Optimizing Spam Email Detection Accuracy Using Advanced Machine Learning Techniques
by Manish Sharma

2025, IJSDR

Key finding: The study investigated the application of multiple machine learning algorithms including Random Forest, Logistic Regression, Naïve Bayes, and SVM across multiple datasets, incorporating feature engineering, bagging, and... Read more
articleView PaperdownloadDownload
keyboard_arrow_downShow more

2. What advances do deep learning and hybrid attention mechanisms offer for detecting spam in email data?

This theme investigates the application of deep learning architectures—especially models integrating convolutional neural networks (CNN), gated recurrent units (GRU), and attention mechanisms—for email spam filtering. These methods focus on hierarchical feature extraction and contextual weighting of informative text segments, aiming to overcome limitations of classical techniques and improve generalization across datasets. The novelty lies in capturing complex semantic structures and temporal dependencies in email content, a crucial advance given the linguistic complexity of spam emails.

Email Spam Detection Using Hierarchical Attention Hybrid Deep Learning Method
by Sultan Zavrak

2023, Email Spam Detection Using Hierarchical Attention Hybrid Deep Learning Method

Key finding: This research proposed a hybrid model combining CNN, GRU, and hierarchical attention mechanisms, which selectively focused on relevant email text parts during training. The temporal convolution layers enabled flexible... Read more
articleView PaperdownloadDownload
A Novel Fuzzy-Logic-Based Multi-Criteria Metric for Performance Evaluation of Spam Email Detection Algorithms
by Rehan Akbar

2023, Applied Sciences

Key finding: The study applied machine learning techniques to identify recurrent word groups characteristic of spam and introduced a feedback-trained model with tokenizers and Naïve Bayes classifiers to distinguish between spam and ham... Read more
articleView PaperdownloadDownload
Context and Machine Learning Based Trust Management Framework for Internet of Vehicles
by Rehan Akbar

2023, Computers, Materials & Continua

Key finding: This research demonstrated the application of neural networks to email spam filtering, showing their capability to learn complex patterns and outperform standard classifiers in accuracy metrics, specifically for phishing and... Read more
articleView PaperdownloadDownload
keyboard_arrow_downShow more

3. How effective are email service providers' pre-acceptance spam filtering techniques and what are the limitations?

This research area delves into the strategies employed by major email providers like Gmail, Yahoo, and Outlook at the SMTP pre-acceptance stage to filter spam, including blacklists, whitelists, and sender reputation analysis. It quantifies the proportion of spam and legitimate emails filtered before message acceptance and analyses the challenges posed by sophisticated spam gangs and end-host spammers. Understanding these filtering boundaries is vital for optimizing server resources and enhancing spam mitigation strategies.

On the Effectiveness of Pre-Acceptance Spam Filtering
by Zhuoqing Mao

2023

Key finding: Through a large-scale empirical study using millions of emails collected at UW-Madison, the authors found that pre-acceptance filtering methods, such as blacklists and whitelists constructed from sender-tracking heuristics,... Read more
articleView PaperdownloadDownload
Machine learning for email spam filtering: review, approaches and open research problems
by Emmanuel Gbenga Dada

2021, Heliyon

Key finding: This review synthesizes state-of-the-art machine learning implementations in major email providers’ spam filters, highlighting Google's advanced neural network-based filtering achieving ~99.9% accuracy. It details innovative... Read more
articleView PaperdownloadDownload
keyboard_arrow_downShow more

Related Topics

  • Spam Filtering
  • Stellar Populations
  • Palmprint Recognition
  • Genetic Algorithm in Optimization
  • Cross-modality
  • Linear Discriminant Analysis
  • Image Annotation
  • Topic modeling
  • Instance-based learning
  • Multi-Objective Evolutionary Algorithms

    • All papers in Email Spam Filter

    Machine learning for email spam filtering: review, approaches and open research problems
    by Adebayo Adetunmbi

    2019, Heliyon

    The upsurge in the volume of unwanted emails called spam has created an intense need for the development of more dependable and robust antispam filters. Machine learning methods of recent are being used to successfully detect and filter... more
    descriptionView Paperarrow_downwardDownload
    Machine learning for email spam filtering: review, approaches and open research problems
    by Stephen B Joseph and 
    1 more
    Emmanuel Gbenga Dada

    2019, Heliyon

    The upsurge in the volume of unwanted emails called spam has created an intense need for the development of more dependable and robust antispam filters. Machine learning methods of recent are being used to successfully detect and filter... more
    descriptionView Paperarrow_downwardDownload
    Critical Analysis of Spam Prevention Techniques
    by Simi Bajaj
    E-mail spam has remained a scourge and menacing nuisance for users, internet and network service operators and providers, in spite of the anti-spam techniques available; and spammers are relentlessly circumventing these anti-spam... more
    descriptionView Paperarrow_downwardDownload
    Machine learning for email spam filtering: review, approaches and open research problems
    by Emmanuel Gbenga Dada

    2019, Heliyon

    The upsurge in the volume of unwanted emails called spam has created an intense need for the development of more dependable and robust antispam filters. Machine learning methods of recent are being used to successfully detect and filter... more
    Fig. 1. The volume of spam emails 4th quarter 2016 to 1st quarter 2018. Many researchers and academicians have proposed different email spam classification techniques which have been successfully used to classify data into groups. These methods include probabilistic, decision
    Fig. 2. Pictorial Representation of the Structure of this paper. There is a rapid increase in the interest being shown by the global research community on email spam filtering. In this section, we presen! similar reviews that have been presented in the literature in this domain. This method is followed so as to articulate the issues that are yet to be addressed and to highlight the differences with our current review. Lueg [17] presented a brief survey to explore the gaps in whether informatior filtering and information retrieval technology can be applied to postulate Email spam detection in a logical, theoretically grounded manner, ir order to facilitate the introduction of spam filtering technique that coulc The rest of this paper is organized as follows: Section 2 gives a
    Fig. 3. Email server spam filtering architecture.
    Fig. 4. Architecture of neural network (NN) Classifier.
    Fig. 5. Rough Set (RS) email filtering process workflow from user mailbox.
    Fig. 6. Decision Tree Algorithm for email spam filtering. (emails contain both spam and ham) of the dataset is reduced. The dataset can be tested using the decision tree algorithm after the tree is created from the training email dataset. The email dataset being tested undergo some processing in the tree using some predefined rules pending the time it will get to a leaf node. The label in the leaf node is then assigned to the tested data. Below in Fig. 6 is a theoretical tree that illustrate how the decision tree algorithm carries out its spam filtering operation. F represents the features or words in the email message. V depicts the values or word frequencies of some words contained in the email message. C depicts the labels which are either spam/ham.
    Summary of previous reviews in email spam filtering. Table 1
    Publicly available email spam corpus. Table 2 on the detection of spam messages solely. In a real world environment where there is nothing like zero probability of wrongly categorizing < ham message, it is required that a compromise be reached between the two kinds of errors, depending on the predisposition of user and the performance indicators used. The formulae for calculating the classifi cation accuracy and classification error are depicted in Eqs. (1) and (2. below: Spam filters with a drastically reduced FPR and FNR are said to have a better performance. These standard characteristics (FNR and FPR) rep- resents the efficiency of filters that directly aim at the classification de- cision borderline devoid of generating the probability estimate. On the other hand, the efficiency of filters that explicitly estimate the group conditional probabilities and then execute classification based on esti- mated probabilities can be represented by a curve called ROC (Receiver Operating Characteristics) curve. ROC curve, is a graphical plot that demonstrates the analytical capability of a spam filter as its bias level is modified [48]. The ROC curve is generated by plotting the true positive rate (TPR) against the false positive rate (FPR) at different threshold settings [49]. The true positive rate is referred to as sensitivity, recall or probability of detection [49] in machine learning. The false-positive rate is referred to as the squabble or likelihood of false alarm. This is computed by subtracting the value of the specificity from 1 (ie. 1 - specificity). ROC testing are an outstanding standard of performance measure in spam filtering [48]. When the ROC curve of a spam filter closely sits on top of another, such filter can be classified a filter with superior performance in all implementation setups [20]. The two metrics imported from the field of information retrieval ‘recall’ and ‘precision’ are respectively utilised for obtaining the efficiency and characteristic of spam filters [50].
    Levels of cost sensitivity of model. Table 3
    Algorithm 1 kNN Algorithm for Spam Email Classification 5.2. Naive Bayes classifier In [58], the steps involved in a simple kNN algorithm for filtering spam mails is described in the algorithm below. Here Neighbours(d) return the k nearest neighbours of d, Closest (d, t) return the closest el- ements of t in d, and testClass(S) return the class label of S. A simple kNN algorithm for spam email classification is in the algorithm below:
    Algorithm 2 Naive Bayes Classification Algorithm for Email Spam Classification The message is classified as spam if the total spamminess product S [M] is greater than the hamminess product H [M]. The above description in [63] is used in the Naive Bayes classification algorithm for email spam classification depicted below:
    Algorithm 5 Email spam classification algorithm using Rough Set
    Algorithm 5 Email spam classification algorithm using Rough Set (continued )
    Algorithm 6 Support Vector Machine (SVM) algorithm 5.7. Decision tree
    Algorithm 7 Decision Tree algorithm for Spam Filtering By partitioning the email dataset in relation to least entropy, the resultant email dataset has the highest information gain and so impurity The decision tree algorithm for classifying email messages using en- tropy algorithm is presented below:
    Algorithm 8 AdaBoost Algorithm for Email Spam Classification (Adapted from [127]) centered on the theory of hybridisation of several weak hypotheses, a very good example is the AdaBoost system. The objective of boosting is to obtain a very accurate classification rule by amalgamating several weak rules or weak hypotheses each of which may be only relatively accurate. A learner is trained in every phase of the classification process, and the result of each phase is used to add credence to data for the upcoming phases [87]. AdaBoost is the most popular boosting algorithm. It was proposed by [88]. AdaBoost can produce a good output even when the performance of the weak learners are unsatisfactory. At present Boosting is now been applied in the field of classification, regression, face recog- nition and so on. Boosting algorithms that utilised confidence rated projections are being applied to solve spam filtering problem. Literature have also shown that they can produce classification results that are better than that of Bayesian and decision tree approaches [87]. AdaBoost has become a widely accepted machine learning algorithm because of its astounding performance in solving classification problems. It is believed among some statisticians that AdaBoost has some relationship with lo- gistic regression probability maximisation [89]. The widespread use of AdaBoost according to Rob Schapire is not unconnected with the ad- vantages that the approach have over some other learning algorithm. AdaBoost is fast, the algorithm is straightforward and easy to program, absence of parameter tuning (except T) makes is less cumbersome. It is adaptable and can combine well with any learning algorithm. Also, there no need of any previous knowledge about weak learner. It is verifiably efficient, provided it can always locate rough rules of thumb. The algo- rithm is very adaptable, and can be used with data that is textual, numeric or discrete in nature. It has been expanded further to learning problems that are outside binary classification. The AdaBoost algorithm for detecting spam email is show in algorithm 8 below:
    Algorithm 9 Random Forests Algorithm for Email Classification
    Algorithm 10 Convolutional Neural Networks for Email Classification
    Summary of published papers that attempted spam filtering using Machine Learning techniques. Table 4
    descriptionView Paperarrow_downwardDownload
    E-Mail Spam Detection using Machine Learning and Deep Learning
    by IJRASET Publication

    2020, international journal for research in applied science and engineering technology ijraset

    Here we present an inclusive review of recent and successful content-based e-mail spam filtering techniques. Our focus is primarily on machine learning-based spam filters and variants that are inspired by them. We report on related... more
    descriptionView Paperarrow_downwardDownload
    Machine learning for email spam filtering: review, approaches and open research problems
    by Shafi'i Muhammad ABDULHAMID and 
    2 more
    Emmanuel Gbenga Dada
    Stephen B Joseph

    2019, Heliyon (ISI & Scopus indexed)

    The upsurge in the volume of unwanted emails called spam has created an intense need for the development of more dependable and robust antispam filters. Machine learning methods of recent are being used to successfully detect and filter... more
    Algorithm 5 Email spam classification algorithm using Rough Set
    Algorithm 5 Email spam classification algorithm using Rough Set (continued )
    Algorithm 6 Support Vector Machine (SVM) algorithm 5.7. Decision tree
    Fig. 1. The volume of spam emails 4th quarter 2016 to 1st quarter 2018. Many researchers and academicians have proposed different email spam classification techniques which have been successfully used to classify data into groups. These methods include probabilistic, decision
    Fig. 2. Pictorial Representation of the Structure of this paper. There is a rapid increase in the interest being shown by the global research community on email spam filtering. In this section, we presen! similar reviews that have been presented in the literature in this domain. This method is followed so as to articulate the issues that are yet to be addressed and to highlight the differences with our current review. Lueg [17] presented a brief survey to explore the gaps in whether informatior filtering and information retrieval technology can be applied to postulate Email spam detection in a logical, theoretically grounded manner, ir order to facilitate the introduction of spam filtering technique that coulc The rest of this paper is organized as follows: Section 2 gives a
    Fig. 3. Email server spam filtering architecture.
    Fig. 4. Architecture of neural network (NN) Classifier.
    Fig. 5. Rough Set (RS) email filtering process workflow from user mailbox.
    Fig. 6. Decision Tree Algorithm for email spam filtering. (emails contain both spam and ham) of the dataset is reduced. The dataset can be tested using the decision tree algorithm after the tree is created from the training email dataset. The email dataset being tested undergo some processing in the tree using some predefined rules pending the time it will get to a leaf node. The label in the leaf node is then assigned to the tested data. Below in Fig. 6 is a theoretical tree that illustrate how the decision tree algorithm carries out its spam filtering operation. F represents the features or words in the email message. V depicts the values or word frequencies of some words contained in the email message. C depicts the labels which are either spam/ham.
    Summary of previous reviews in email spam filtering. Table 1
    Publicly available email spam corpus. Table 2 on the detection of spam messages solely. In a real world environment where there is nothing like zero probability of wrongly categorizing < ham message, it is required that a compromise be reached between the two kinds of errors, depending on the predisposition of user and the performance indicators used. The formulae for calculating the classifi cation accuracy and classification error are depicted in Eqs. (1) and (2. below: Spam filters with a drastically reduced FPR and FNR are said to have a better performance. These standard characteristics (FNR and FPR) rep- resents the efficiency of filters that directly aim at the classification de- cision borderline devoid of generating the probability estimate. On the other hand, the efficiency of filters that explicitly estimate the group conditional probabilities and then execute classification based on esti- mated probabilities can be represented by a curve called ROC (Receiver Operating Characteristics) curve. ROC curve, is a graphical plot that demonstrates the analytical capability of a spam filter as its bias level is modified [48]. The ROC curve is generated by plotting the true positive rate (TPR) against the false positive rate (FPR) at different threshold settings [49]. The true positive rate is referred to as sensitivity, recall or probability of detection [49] in machine learning. The false-positive rate is referred to as the squabble or likelihood of false alarm. This is computed by subtracting the value of the specificity from 1 (ie. 1 - specificity). ROC testing are an outstanding standard of performance measure in spam filtering [48]. When the ROC curve of a spam filter closely sits on top of another, such filter can be classified a filter with superior performance in all implementation setups [20]. The two metrics imported from the field of information retrieval ‘recall’ and ‘precision’ are respectively utilised for obtaining the efficiency and characteristic of spam filters [50].
    Levels of cost sensitivity of model. Table 3
    Algorithm 1 kNN Algorithm for Spam Email Classification 5.2. Naive Bayes classifier In [58], the steps involved in a simple kNN algorithm for filtering spam mails is described in the algorithm below. Here Neighbours(d) return the k nearest neighbours of d, Closest (d, t) return the closest el- ements of t in d, and testClass(S) return the class label of S. A simple kNN algorithm for spam email classification is in the algorithm below:
    Algorithm 2 Naive Bayes Classification Algorithm for Email Spam Classification The message is classified as spam if the total spamminess product S [M] is greater than the hamminess product H [M]. The above description in [63] is used in the Naive Bayes classification algorithm for email spam classification depicted below:
    Algorithm 7 Decision Tree algorithm for Spam Filtering By partitioning the email dataset in relation to least entropy, the resultant email dataset has the highest information gain and so impurity The decision tree algorithm for classifying email messages using en- tropy algorithm is presented below:
    Algorithm 8 AdaBoost Algorithm for Email Spam Classification (Adapted from [127]) centered on the theory of hybridisation of several weak hypotheses, a very good example is the AdaBoost system. The objective of boosting is to obtain a very accurate classification rule by amalgamating several weak rules or weak hypotheses each of which may be only relatively accurate. A learner is trained in every phase of the classification process, and the result of each phase is used to add credence to data for the upcoming phases [87]. AdaBoost is the most popular boosting algorithm. It was proposed by [88]. AdaBoost can produce a good output even when the performance of the weak learners are unsatisfactory. At present Boosting is now been applied in the field of classification, regression, face recog- nition and so on. Boosting algorithms that utilised confidence rated projections are being applied to solve spam filtering problem. Literature have also shown that they can produce classification results that are better than that of Bayesian and decision tree approaches [87]. AdaBoost has become a widely accepted machine learning algorithm because of its astounding performance in solving classification problems. It is believed among some statisticians that AdaBoost has some relationship with lo- gistic regression probability maximisation [89]. The widespread use of AdaBoost according to Rob Schapire is not unconnected with the ad- vantages that the approach have over some other learning algorithm. AdaBoost is fast, the algorithm is straightforward and easy to program, absence of parameter tuning (except T) makes is less cumbersome. It is adaptable and can combine well with any learning algorithm. Also, there no need of any previous knowledge about weak learner. It is verifiably efficient, provided it can always locate rough rules of thumb. The algo- rithm is very adaptable, and can be used with data that is textual, numeric or discrete in nature. It has been expanded further to learning problems that are outside binary classification. The AdaBoost algorithm for detecting spam email is show in algorithm 8 below:
    Algorithm 9 Random Forests Algorithm for Email Classification
    Algorithm 10 Convolutional Neural Networks for Email Classification
    Summary of published papers that attempted spam filtering using Machine Learning techniques. Table 4
    descriptionView Paperarrow_downwardDownload
    Detection of Email Spam using Natural Language Processing Based Random Forest Approach
    by IJCSMC Journal and 
    1 more
    Nivedha Arasu

    2022, IJCSMC

    An unsolicited means of digital communications in the internet world is the spam email, which could be sent to an individual or a group of individuals or a company. These spam emails may cause serious threat to the user i.e., the email... more
    descriptionView Paperarrow_downwardDownload
    Whale optimization algorithm-based email spam feature selection method using rotation forest algorithm for classification
    by Shafi'i Muhammad ABDULHAMID and 
    1 more
    Maryam Shuaib

    2019, SN Applied Sciences

    Email has continued to be an integral part of our lives and as a means for successful communication on the internet. The problem of spam mails occupying a huge amount of space and bandwidth, and the weaknesses of spam filtering techniques... more
    Fig. 2. Pseudocode of the rotation forest ensemble method
    It indicates the number of instances which are positively classified and are relevant. A high precision shows high rel- evance in detecting positives.
    The accuracy is used to show the level of correct predic- tions. The value 1 is largest, indicating the highest accu- racy, when the rotation forest classification algorithm was run on the Spambase dataset after feature selection with
    WOA-rotation forest recorded a high F-measure of 0.9990 and 0.9944 for Spambase and Enron datasets, respectively, and this is seen in Fig. 8. for the Enron dataset. Figure 10 illustrates the root mean squared error. The kappa characteristic gives the level of agreements between the true classes and the classifications. The value 1 is highest showing total agreement; in this study, Spambase dataset showed a high kappa characteristics of 0.9971 which was got when the test was carried out with 20-fold cross-validation, while Enron dataset also with 20-fold cross-validation had 0.9854; Fig. 9 shows the respective kappa characteristics for the three test options used. 6.1.2 Performance comparison before and after feature selection The test was carried out on the datasets before feature selection with the proposed WOA and after feature selec- tion. There is a significant increase in accuracy from 94.2 to 99.89% for the Spambase dataset with a drop in FP rate from 0.067 to 0.0019. Enron dataset recorded improved accuracy from 96.9 to 99.43% and a fall in FP rate from 0.302 to 0.007. The results of the experiment which shows a performance improvement in the metrics are shown in Fig. 11 and Fig. 12 in that other.
    proposed the use of WOA to select salient features in the email corpus and rotation forest algorithm for classifying emails as spam and non-spam. In achieving the aim and objectives of this research, the Spambase dataset from the UCI repository with 58 attributes and 4601 instances (spam and non-spam emails) and the Enron-Spam corpus were used. The entire datasets were used, and the evaluation of the rotation forest algorithm was done before and after feature selection with WOA using 10-fold cross-valida- tion, 20-fold cross-validation and 66% split test options. The rotation forest algorithm after feature selection with WOA was able to classify the emails into spam and non-spam with a performance accuracy of 99.89% and a low FP rate of 0.0019. The result obtained hence shows clearly that after feature selection with WOA, the rotation
    forest algorithm outperformed 98.4% [38], 97.2% [13] and 76% [21]. forest algorithm outperformed 98.4% [38], 97.2% [13] and 76% [21].
    Table 1 Attribute description of the Soambase dataset In order to adequately classify the email spam corpus, ten- fold cross-validation was used because the larger the sam- ple used for training, the better the performance of the classifier, but the returns start to decrease once a particular amount of training data are surpassed. Also the larger the testing sample, the higher the accuracy for the estimation
    Table 2 Summary of results after feature selection with WOA on Spambase dataset using 10-fold cross-validation Table 3 Summary of results after feature selection with WOA on Spambase dataset using 20-fold cross-validation
    Table 4 Summary of results after feature selection with WOA on Spambase dataset using 66% split
    WOA, 99.89% accuracy was achieved with 10-fold and 20-fold cross-validation and a slight drop to 99.77% when 66% spilt was used. The Enron dataset also displayed a high accuracy of 99.3% with 20-fold cross-validation and 98.2% when run with 66% split test option. Figure 6 shows the accuracy for the different test options.
    Table5 Summary of results after feature selection with WOA on Enron dataset using 10-fold cross-validation Table6 Summary of results after feature selection with WOA on Enron dataset using 20-fold cross-validation
    Table 7 Summary of results after feature selection with WOA on Enron dataset using 66% Split cross-validation for the Spambase dataset and 0.007 using 20-fold cross-validation for the Enron dataset.
    descriptionView Paperarrow_downwardDownload
    PENGENALAN DAN PENCEGAHAN EMAIL SPAM
    by Raka I Q B A L Syamsuddin
    Email (Elektronik Mail) atau surat elektronik merupakan salah satu perkembangan teknologi saat ini, dengan email pengiriman pesan dapat dilakukan dengan cepat, dan dapat dikirimkan ke banyak penerima pesan dalam waktu yang singkat, Namun... more
    descriptionView Paperarrow_downwardDownload
    E-Mail Spam Detection using Machine Learning and Deep Learning
    by Shivam pandey

    International Journal for Research in Applied Science and Engineering Technology

    Here we present an inclusive review of recent and successful content-based e-mail spam filtering techniques. Our focus is primarily on machine learning-based spam filters and variants that are inspired by them. We report on related ideas,... more
    descriptionView Paperarrow_downwardDownload
    Spamizer : An approach to handle web form spam
    by Dr. Manish Saxena
    –The Spam Emails are regularly causing huge losses to business on a regular basis. The Spam filtering is an automated technique to identity SPAM and HAM (Non-Spam). The Web Spam filters can be categorized as: Content based spam filters... more
    Fig. 1. Anti-Spam strategies- Detection —based, Demotion-based and Prevention based strategies. [4] computer. These spam messages are quite cheap to send such information’s and even if one in a thousand spam recipient will respond to such messages, the spam sender will be in huge profit.[2]. The Spam filtering is an automated technique to identity SPAM and Non-Spam also known as HAM. Generally on the basis of message contents Spam filter is taking its decision about SPAM / HAM, on basis of sender and receiver’s characteristics. Spam filters are having knowledge or experience that if other users have reported similar messages as Spam or not. Such Spam filters are not so perfect, like us, therefore we need to change the working of Spam Filters by adding some more constraints to their techniques.
    Approach-2: Spam Prevention using Hidden Form Field It is not possible for a Spambot to distinguish between optional or mandatory web form fields, therefore a spambot just fill every field on a web form. Every website is having different style to mark such mandatory fields, someone them are using asterisks on such required fields, some are using red colored fonts, some are using HTML5S attribute “required”, and some don’t bother at all to mark such fields, but they redirect you back to the web form if you have missed one such field.
    You may make it more complex by giving the field a fleid ID or a class name which would then force the bot to scan through your CSS files to determine the element’s visibility [7]. Fig. 3. Spamizer Block Diagram- We have implemented Spamizer as a Web Service which is managing number of contact forms on different web pages
    VII. LIMITATIONS OF THE CURRENT WORK
    TABLE I. Opservep spam STATISTICS
    descriptionView Paperarrow_downwardDownload
    Semi-supervised novelty detection with one class SVM for SMS spam detection
    by Suleiman Yerima

    2022, The 29th International Conference on Systems, Signals and Image Processing, IWSSIP 2022 , June 1-3, Sofia, Bulgaria

    The volume of SMS messages sent on a daily basis globally has continued to grow significantly over the past years. Hence, mobile phones are becoming increasingly vulnerable to SMS spam messages, thereby exposing users to the risk of fraud... more
    descriptionView Paperarrow_downwardDownload
    Can we CAN the Email Spam
    by Simi Bajaj
    The purpose of email spam is to advertise to sell, phishing attacks, DDOS attacks and many more. Many solutions of various kinds such as blacklisting, whitelisting, grey-listing, content filtering have been proposed at the sender and... more
    descriptionView Paperarrow_downwardDownload
    Highly Discriminative Statistical Features for Email Classification
    by Juan Carlos Gomez and 
    1 more
    Erik Boiy
    This paper reports on email classification and filtering, more specifically on spam versus ham and phishing versus spam classification, based on content features. We test the validity of several novel statistical feature extraction... more
    descriptionView Paperarrow_downwardDownload
    Taxonomy and Control Measures of SPAM and SPIM
    by Simi Bajaj
    In this age of electronic money transactions, the opportunities for electronic crime expanded at the same rate as ever expanding rise of on-line services. With world becoming a global village, crime over the internet transcends no... more
    The problem of Spam in mobile devices is twofold. While the problem of spam may not bother users so much as on a networked PC, it really affects them on small mobile devices for the reasons of cost, time consumption and inconvenience. If they receive up to 15 spam mails via GPRS or UMTS it does cost real money deleting and getting rid of it on small screen devices. Apart from the usual unsolicited spam messages in inbox, mobile devices also receiving unwanted, unsolicited Instant messages (SMS and MMS). Such Spam in mobile devices is called SPIM (Spam thru Instant Messages). According to report from Adaptive Mobile [7], survey done on smart phone (mobile device) users in UK (1000 participants) in May 2011 reported that 69% received SMS text phishing and 66% SMS spam. In Europe and Asian countries SMS spam is a fashion generating almost half of the total SMS traffic in some countries [8]. The number has been increasing since then [9] reaching 15-25 messages per day . This practice is more common in Asian countries due to poverty and it is easier to lure people to carry out tasks with a lure to make money. The conversion rate of spam sent compared to products bought is of prime importance in driving the need to reduce spam in mobile devices [10]. Since users of smart phones expose themselves to security risks as reported in [7], 50% would open an SMS text message from someone they don’t know, 36% would open an email on their mobile from someone they don’t know and 32% save log-in information such as passwords to their mobile, it is crucial to address the problem of spam in mobile devices.
    could lure user to a malicious site or convince them to install malicious code on your portable device [18]. In any phishing (type of spam- identity theft) attack first few hours are very crucial as many attacks are blocked or the site are taken down after that. The chances that a mobile device user will be hit are much higher than a desktop user, since these spam emails arrive on the mobile users device first and they are 3 times more likely to submit their login details than the desktop users [19].
    descriptionView Paperarrow_downwardDownload
    Whale optimization algorithm-based email spam feature selection method using rotation forest algorithm for classification
    by Ismaila Idris

    2019, SN Applied Sciences

    Email has continued to be an integral part of our lives and as a means for successful communication on the internet. The problem of spam mails occupying a huge amount of space and bandwidth, and the weaknesses of spam filtering techniques... more
    descriptionView Paperarrow_downwardDownload
    Must-Read blockchain Articles in 2019
    by International Journal of Security, Privacy and Trust Management (IJSPTM)

    2019

    Ethereum is an open-source, public, block chain-based distributed computing platform and operating system featuring smart contract functionality. In this paper, we proposed an Ethereum based electronic voting (e-voting) protocol,... more
    descriptionView Paperarrow_downwardDownload
    Bayesov spam filter
    by Dino Kliček

    2019, Fakultet organizacije i informatike, Varaždin

    Cilj ovog rada je izgraditi anti-spam filter, softverski alati koji automatski prepoznaje dolazne neželjene poruke. Dakle, radi se o machine learning sustavu koji na temelju naučenih podataka o tome što je dobar mail, a što loš može... more
    descriptionView Paperarrow_downwardDownload
    Identifying and Resolving Hidden Text Salting
    by Juan Carlos Gomez and 
    1 more
    Erik Boiy
    Hidden salting in digital media involves the intentional addition or distortion of content patterns with the purpose of content filtering. We propose a method to detect portions of a digital text source which are invisible to the end... more
    descriptionView Paperarrow_downwardDownload
    Machine learning for email spam filtering: review, approaches and open research problems
    by Emmanuel Gbenga Dada

    2019, Heliyon

    The upsurge in the volume of unwanted emails called spam has created an intense need for the development of more dependable and robust antispam filters. Machine learning methods of recent are being used to successfully detect and filter... more
    Fig. 1. The volume of spam emails 4th quarter 2016 to 1st quarter 2018. Many researchers and academicians have proposed different email spam classification techniques which have been successfully used to classify data into groups. These methods include probabilistic, decision
    Fig. 2. Pictorial Representation of the Structure of this paper. There is a rapid increase in the interest being shown by the global research community on email spam filtering. In this section, we presen! similar reviews that have been presented in the literature in this domain. This method is followed so as to articulate the issues that are yet to be addressed and to highlight the differences with our current review. Lueg [17] presented a brief survey to explore the gaps in whether informatior filtering and information retrieval technology can be applied to postulate Email spam detection in a logical, theoretically grounded manner, ir order to facilitate the introduction of spam filtering technique that coulc The rest of this paper is organized as follows: Section 2 gives a
    Fig. 3. Email server spam filtering architecture.
    Fig. 4. Architecture of neural network (NN) Classifier.
    Fig. 5. Rough Set (RS) email filtering process workflow from user mailbox.
    Fig. 6. Decision Tree Algorithm for email spam filtering. (emails contain both spam and ham) of the dataset is reduced. The dataset can be tested using the decision tree algorithm after the tree is created from the training email dataset. The email dataset being tested undergo some processing in the tree using some predefined rules pending the time it will get to a leaf node. The label in the leaf node is then assigned to the tested data. Below in Fig. 6 is a theoretical tree that illustrate how the decision tree algorithm carries out its spam filtering operation. F represents the features or words in the email message. V depicts the values or word frequencies of some words contained in the email message. C depicts the labels which are either spam/ham.
    Summary of previous reviews in email spam filtering. Table 1
    Publicly available email spam corpus. Table 2 on the detection of spam messages solely. In a real world environment where there is nothing like zero probability of wrongly categorizing < ham message, it is required that a compromise be reached between the two kinds of errors, depending on the predisposition of user and the performance indicators used. The formulae for calculating the classifi cation accuracy and classification error are depicted in Eqs. (1) and (2. below: Spam filters with a drastically reduced FPR and FNR are said to have a better performance. These standard characteristics (FNR and FPR) rep- resents the efficiency of filters that directly aim at the classification de- cision borderline devoid of generating the probability estimate. On the other hand, the efficiency of filters that explicitly estimate the group conditional probabilities and then execute classification based on esti- mated probabilities can be represented by a curve called ROC (Receiver Operating Characteristics) curve. ROC curve, is a graphical plot that demonstrates the analytical capability of a spam filter as its bias level is modified [48]. The ROC curve is generated by plotting the true positive rate (TPR) against the false positive rate (FPR) at different threshold settings [49]. The true positive rate is referred to as sensitivity, recall or probability of detection [49] in machine learning. The false-positive rate is referred to as the squabble or likelihood of false alarm. This is computed by subtracting the value of the specificity from 1 (ie. 1 - specificity). ROC testing are an outstanding standard of performance measure in spam filtering [48]. When the ROC curve of a spam filter closely sits on top of another, such filter can be classified a filter with superior performance in all implementation setups [20]. The two metrics imported from the field of information retrieval ‘recall’ and ‘precision’ are respectively utilised for obtaining the efficiency and characteristic of spam filters [50].
    Levels of cost sensitivity of model. Table 3
    Algorithm 1 kNN Algorithm for Spam Email Classification 5.2. Naive Bayes classifier In [58], the steps involved in a simple kNN algorithm for filtering spam mails is described in the algorithm below. Here Neighbours(d) return the k nearest neighbours of d, Closest (d, t) return the closest el- ements of t in d, and testClass(S) return the class label of S. A simple kNN algorithm for spam email classification is in the algorithm below:
    Algorithm 2 Naive Bayes Classification Algorithm for Email Spam Classification The message is classified as spam if the total spamminess product S [M] is greater than the hamminess product H [M]. The above description in [63] is used in the Naive Bayes classification algorithm for email spam classification depicted below:
    Algorithm 5 Email spam classification algorithm using Rough Set
    Algorithm 5 Email spam classification algorithm using Rough Set (continued )
    Algorithm 6 Support Vector Machine (SVM) algorithm 5.7. Decision tree
    Algorithm 7 Decision Tree algorithm for Spam Filtering By partitioning the email dataset in relation to least entropy, the resultant email dataset has the highest information gain and so impurity The decision tree algorithm for classifying email messages using en- tropy algorithm is presented below:
    Algorithm 8 AdaBoost Algorithm for Email Spam Classification (Adapted from [127]) centered on the theory of hybridisation of several weak hypotheses, a very good example is the AdaBoost system. The objective of boosting is to obtain a very accurate classification rule by amalgamating several weak rules or weak hypotheses each of which may be only relatively accurate. A learner is trained in every phase of the classification process, and the result of each phase is used to add credence to data for the upcoming phases [87]. AdaBoost is the most popular boosting algorithm. It was proposed by [88]. AdaBoost can produce a good output even when the performance of the weak learners are unsatisfactory. At present Boosting is now been applied in the field of classification, regression, face recog- nition and so on. Boosting algorithms that utilised confidence rated projections are being applied to solve spam filtering problem. Literature have also shown that they can produce classification results that are better than that of Bayesian and decision tree approaches [87]. AdaBoost has become a widely accepted machine learning algorithm because of its astounding performance in solving classification problems. It is believed among some statisticians that AdaBoost has some relationship with lo- gistic regression probability maximisation [89]. The widespread use of AdaBoost according to Rob Schapire is not unconnected with the ad- vantages that the approach have over some other learning algorithm. AdaBoost is fast, the algorithm is straightforward and easy to program, absence of parameter tuning (except T) makes is less cumbersome. It is adaptable and can combine well with any learning algorithm. Also, there no need of any previous knowledge about weak learner. It is verifiably efficient, provided it can always locate rough rules of thumb. The algo- rithm is very adaptable, and can be used with data that is textual, numeric or discrete in nature. It has been expanded further to learning problems that are outside binary classification. The AdaBoost algorithm for detecting spam email is show in algorithm 8 below:
    Algorithm 9 Random Forests Algorithm for Email Classification
    Algorithm 10 Convolutional Neural Networks for Email Classification
    Summary of published papers that attempted spam filtering using Machine Learning techniques. Table 4
    descriptionView Paperarrow_downwardDownload
    Using message features and sender identity for email spam filtering
    by Jay Buckingham

    2011, United States Patent

    Email spam filtering is performed based on a sender reputation and message features. When an email message is received, a preliminary spam determination is made based, at least in part, on a combination of a reputation associated with the... more
    descriptionView Paperarrow_downwardDownload
    PCA Document Reconstruction for Email Classification
    by Juan Carlos Gomez
    This paper presents a document classifier based on text content features and its application to email classification. We test the validity of a classifier which uses Principal Component Analysis Document Reconstruction (PCADR), where the... more
    descriptionView Paperarrow_downwardDownload
    Anticipating Hidden Text Salting in Emails
    by Patrick Horkan

    2008, Lecture Notes in Computer Science

    Salting is the intentional addition or distortion of content, aimed to evade automatic filtering. Salting is usually found in spam emails. Salting can also be hidden in phishing emails, which aim to steal personal information from users.... more
    descriptionView Paperarrow_downwardDownload
    Email Spam Detection Generation Algorithm for Negative Selection Algorithm with Hamming Distance Partial Matching Rules
    by Shafi'i Muhammad ABDULHAMID

    2014, Australian Journal of Basic and Applied Sciences (Thomson Reuters Master List indexed)

    Negative selection algorithms (NSAs) are inspired by artificial immune system. It creates techniques that aim at developing the immune based model. This is done by distinguishing self from non-self spam in the generation of detectors. In... more
    descriptionView Paperarrow_downwardDownload
    Whale optimization algorithm-based email spam feature selection method using rotation forest algorithm for classification
    by Olawale Adebayo

    SN Applied Sciences

    Email has continued to be an integral part of our lives and as a means for successful communication on the internet. The problem of spam mails occupying a huge amount of space and bandwidth, and the weaknesses of spam filtering techniques... more
    descriptionView Paperarrow_downwardDownload
    Whale optimization algorithm-based email spam feature selection method using rotation forest algorithm for classification
    by John K. Alhassan

    SN Applied Sciences

    Email has continued to be an integral part of our lives and as a means for successful communication on the internet. The problem of spam mails occupying a huge amount of space and bandwidth, and the weaknesses of spam filtering techniques... more
    descriptionView Paperarrow_downwardDownload
    Using Biased Discriminant Analysis for Email Filtering
    by Juan Carlos Gomez and 
    1 more
    M. Moens
    This paper reports on email filtering based on content features. We test the validity of a novel statistical feature extraction method, which relies on dimensionality reduction to retain the most informative and discriminative features... more
    descriptionView Paperarrow_downwardDownload
    Download research papers for free!
    580 California St., Suite 400
    San Francisco, CA, 94104