Interestingness Measures

Je remercie en premier lieu Monsieur André TOTOHASINA, Professeur titulaire de l'Université, pour m'avoir proposé ce sujet, et m'avoir supporté dans toutes les étapes de cette thèse. Ses avis, ses nombreux conseils et son soutien constant m'ont permis de m'améliorer sur de nombreux points. Qu'il trouve ici l'expression de mon plus grand respect et qu'il reçoive mes plus vifs remerciements. Ce travail de thèse a été en partie réalisé dans le cadre d'un projet collaboratif, qui m'a permis d'explorer des domaines de recherche connexes à ma thèse et d'établir des collaborations avec d'autres chercheurs. Je remercie également Monsieur Daniel Rajaonasy FENO, Maître de conférences, d'accepter d'être co-encadreur de cette thèse. Ses idées, ses multitudes directives et son soutien moral permanent m'ont donné des inspirations afin que je puisse réaliser le présent travail jusqu'à la fin. Il m'a fourni satisfaction à tout moment opportun. Ledit travail de thèse est étroitement réalisé en sa collaboration, ce qui m'a permis d'évoluer mes connaissances car la collaboration mentionnée a servi de guide fiable, menant ainsi à l'objectif souhaité sans trébucher. J'adresse particulièrement mes remerciements et mes sincères respects aux membres du Jury pour la qualité de leurs jugements très pertinents et leurs remarques constructives sur ma thèse. Je remercie Monsieur RAHERINIRINA Angelo Fulgence, Maître de conférences HDR et Monsieur REGNIER Jean-Claude, Professeur Emérite des universités d'avoir accepté d'être rapporteurs de mes travaux. Je profite cette occasion pour leur adresser mes sincères respects et leur exprimer ma profonde gratitude pour leurs commentaires et jugements très pertinents et constructifs sur ma thèse, tant sur le fond que sur la forme . « Ny hazo no vanon-ko lakana, ny tany naniriany no tsara» ("Si l'on peut construire une pirogue à partir d'un arbre, c'est parce que le terroir où il a poussé est fertile"). Mes reconnaissances les plus sincères vont : aux enseignants et le dirigeant de l'Ecole Normale Supérieure pour l'enseignement Technique Antsiranana (ENSET), aux enseignants et le dirigeant de Lycée Mixtes Nosy Be, aux enseignants et le dirigeant de l'Ecole Privée MIZPA Nosy Be, aux enseignants et le dirigeant de l'EPP Andavakotoko Nosy Be. J'exprime ainsi toute ma reconnaissance à Monsieur Armand, Docteur de l'Université de Toamasina et mon collègue chercheur, pour les échanges scientifiques et techniques que nous avons tenus. Je remercie mille fois mon Père Masonova, l'un des principaux financiers de travail après vi moi même qu'il a transmis de conseil de travailler dur pour pouvoir de s'amuser après. Je me souviens toujours notre règle que je respecte jusqu'aujourd'hui. Je vous remercie pour cela. A la mémoire de ma mère adoptive ZAFIMONINA qui m'a élevée depuis l'âges de 5ans, et m'a donné une véritable amour jusqu'à ce que j'ai grandi. Et à ma mère biologique Geogine ADBOUDOU qui est toujours là pour moi, je vous remercie sincèrement pour cela. Enfin, mes reconnaissances les plus sincères vont à l'endroit de mon épouse RASOANI-RINA Sabrina, ainsi qu'à l'ensemble de ma famille BEMENA, plus particulièrement mes soeurs Prisca, Tinah, et ma cousine Angel Patricia ZAFIMAMY qui m'ont toujours soutenu dans tout ce que j'ai entrepris. Il m'est impossible d'imaginer en être arrivé au stade actuel sans elles. vii Dédicace ii Avant-propos iii Remerciements vi Notations et abréviations xv R R ∪ [-∞ ; ∞] s Taux susceptibilité (ou de transmission des retirés aux susceptibles ) S Série Scientifique S(t) Ensemble d'individus susceptibles à l'instant t SEI Modèle

Social media platforms have become an important part of our daily lives due to the widespread use of the Internet. They contains a great wealth of valuable information which provide opportunities for us to explore hidden patterns or unknown correlation relationships. The most existing  Association  Rule Mining (ARM)  algorithms  focus  on  mining  associative  patterns  from  large sentiment datasets  based on traditional support-confidence framework. Those algorithms generate a large number of redundant patterns, the majority of which are uninteresting to the user or do not imply a true correlation relationship between related items. In  this  paper,  an  effective  approach  that  integrates  Natural  Languages  Processing (NLP)  techniques, and  ARM concepts in  one  vertical  mining approach in order to discover interesting sentiment associative patterns  is  proposed.  The  proposed  approach  uses  common  NLP  techniques  as  a pre-processing  phase  to  generate  stemmed  wordsets  from  large  sentiment  datasets. In the second phase, the proposed approach pushes the support-all-confidence interestingness measures as a new framework deep during the mining process in order to generate a reduced and complete set of All-Confident Frequent Sentiment Wordsets (ACFSWs) directly from large vertical sentiment datasets.  Furthermore, the generated ACFSWs are used as an underlying knowledge representation for Interesting Sentiment Association Rules (ISARs) discovery. The obtained results show the usefulness and effectiveness of the proposed approach.

In the literature, the properties of several interestingness measures have been analyzed and a framework has been proposed for selecting a right interestingness measure for extracting association rules. As rare association rules contain useful knowledge, researchers are making efforts to investigate efficient approaches to extract the same. In this paper, we make an effort to analyze the properties of interestingness measures for determining the interestingness of rare association rules. Based on the analysis, we suggest a set of properties a user should consider while selecting a measure to find the interestingness of rare associations. The experiments on real-world datasets show that the measures that satisfy the suggested properties can determine the interestingness of rare association rules.

L'extraction de connaissances à partir de données a pour objet la découverte de connaissances à partir de grandes quantités de données, par des méthodes d'apprentissage automatiques ou semi-automatiques, et l'utilisation industrielle ou opérationnelle de ces connaissances. Nous explorons ici une application concrète de la fouille de données, sur un système de serveur vocal. Nous nous intéressons en particulier à l'étape cruciale de validation de motifs extraits via l'utilisation de mesures de qualité.

La mesure de la qualité des connaissances est uneétape clef d'un processus de découverte de règles d'association. Dans cet article, nous présentons IPEE, un indice de qualité de règle qui a la particularité unique d'associer les deux caractéristiques suivantes : d'une part, il est fondé sur un modèle probabiliste, et d'autre part, il mesure unécartà l'équilibre (incertitude maximum de la conclusion sachant la prémisse vraie).

Scientific datasets often consist of complex data types such as images. Mining such data presents interesting issues related to semantics. In this paper, we explore the research issues in mining data from the field of nanotechnology. More specifically, we focus on a problem that relates to image comparison of material nanostructures. A significant challenge here relates to the notion of similarity between the images. Features such as size and height of nano-particles and inter-particle distance are important in image similarity as conveyed by domain experts. However, there are no precise notions of similarity defined apriori. Hence there is a need for learning similarity measures. In this paper, we describe our proposed approach to learn similarity measures for graphical data. We discuss this with reference to nanostructure images. Other challenges in image comparison are also outlined. The use of this research is discussed with respect to targeted applications.

Pattern mining usually results in huge amounts of patterns, among which only small percentages are interesting. In this paper, Surprisingness (including Surpringness_I and Surpringness_II) is proposed as an innovative objective multivariate interestingness measure for automatically identifying interesting patterns from a large quantity of patterns. Surprisingness is applicable in unstructured or semi-structured, multi-domain or mixed-domain data compared to existing measures. An experiment has been conducted enabling unsupervised learning of common sense, interesting patterns and exceptions from a knowledge graph database built from Wikipedia 1 extracted data (represented as directed labeled hypergraphs), using Surpringness.

Due to the increase in data mining research and applications, selection of interesting rules among a huge number of learned rules is an important task in data mining applications. In this paper, the metrics for the interestingness of a rule is investigated and an algorithm that can classify the learned rules according to their interestingness is developed. Classification algorithms were designed to maximize the number of correctly classified instances, given a set of unseen test cases. Furthermore, feature projection based classification algorithms were tested and shown to be successful in large number of real domains. So, in this work, a feature projection based classification algorithm (VFI, Voting Feature Intervals) is adapted to the rule interestingness problem, and FPRC (Feature Projection Based Rule Classification) algorithm is developed.

In most of the real-world domains, benefit and costs of classifications can be dependent on the characteristics of individual examples. In such cases, there is no static benefit matrix available in the domain and each classification benefit is calculated separately. This situation, called feature dependency, is evaluated in the framework of our newly proposed classification algorithm Benefit Maximizing classifier with Feature Intervals (BMFI) that uses feature projection based knowledge representation. This new approach has been evaluated over bank loan applications and experimental results are presented.

Due to the increase in data mining research and applications, selection of interesting rules among a huge number of learned rules is an important task in data mining applications. In this paper, the metrics for the interestingness of a rule is investigated and an algorithm that can classify the learned rules according to their interestingness is developed. Classification algorithms were designed to maximize the number of correctly classified instances, given a set of unseen test cases. Furthermore, feature projection based classification algorithms were tested and shown to be successful in large number of real domains. So, in this work, a feature projection based classification algorithm (VFI, Voting Feature Intervals) is adapted to the rule interestingness problem, and FPRC (Feature Projection Based Rule Classification) algorithm is developed.

In a typical application of association rule learning from market basket data, a set of transactions for a fixed period of time is used as input to rule learning algorithms. For example, the well-known Apriori algorithm can be applied to learn a set of association rules from such a transaction set. However, learning association rules from a set of transactions is not a one time only process. For example, a market manager may perform the association rule learning process once every month over the set of transactions collected through the last month. For this reason, we will consider the problem where transaction sets are input to the system as a stream of packages. The sets of transactions may come in varying sizes and in varying periods. Once a set of transactions arrive, the association rule learning algorithm is executed on the last set of transactions, resulting in new association rules. Therefore, the set of association rules learned will accumulate and increase in number over time, making the mining of interesting ones out of this enlarging set of association rules impractical for human experts. We refer to this sequence of rules as ''association rule set stream" or ''streaming association rules" and the main motivation behind this research is to develop a technique to overcome the interesting rule selection problem. A successful association rule mining system should select and present only the interesting rules to the domain experts. However, definition of interestingness of association rules on a given domain usually differs from one expert to another and also over time for a given expert. This paper proposes a post-processing method to learn a subjective model for the interestingness concept description of the streaming association rules. The uniqueness of the proposed method is its ability to formulate the interestingness issue of association rules as a benefit-maximizing classification problem and obtain a different interestingness model for each user. In this new classification scheme, the determining features are the selective objective interestingness factors related to the interestingness of the association rules, and the target feature is the interestingness label of those rules. The proposed method works incrementally and employs user interactivity at a certain level. It is evaluated on a real market dataset. The results show that the model can successfully select the interesting ones.

Latar Belakang : Asma bronkhial secara umum terjadi ketika bronkhi mengalami inflamasi atau peradangan dan respon berlebih akibat suatu rangsangan yang menyebabkan penyempitan pada saluran pernapasan. Asma bronkhial ditandai dengan kesulitan bernapas atau sesak napas, batuk, mengi, dan meningkatnya produksi sputum. Banyak faktor penyebab terjadinya asma, seperti polusi udara, dingin, stress, dan kecapekan. Hal ini membuat penyakit asma masih banyak dijumpai di wilayah kartasura. Dari data Puskesmas angka kejadianya sebanyak 39 kasus selama bulan April 2013. Tujuan : Untuk megetahui asuhan keperawatan keluarga pada pasien dengan asma meliputi pengkajian, intervensi keperawatan, implementasi dan evaluasi Hasil : Setelah dilakukan penkajian dan asuhan keperawatan keluarga selama 3x kunjungan dalam satu minggu di dapatkan hasil pasien terlihat sesak napas, batuk, suara napas mengi, keluarga tidak bisa menjelaskan tentang penyakitnya, dan setelah dilakukan tindakan keperawatan bersihan j...

Generalized rule discovery is a rule discovery framework that subsumes association rule discovery and the type of search employed to find individual rules in classification rule discovery. This new rule discovery framework escapes the limitations of the support-confidence framework inherent in association rule discovery. This empowers data miners to identify the types of rules that they wish to discover and develop efficient algorithms for discovering those rules. This paper presents a scalable algorithm applicable to a wide range of generalized rule discovery task and demonstrates its efficiency.

In data mining it is usually desirable that discovered knowledge have some characteristics such as being as accurate as possible, comprehensible and surprising to the user. The vast majority of data mining algorithms produce, as part of their results, information of a statistical nature that allows the user to assess how accurate and reliable the discovered knowledge is. However, in many cases this is not enough for the user. Even if discovered knowledge is highly accurate from a statistical point of view, it might not be interesting for the user. Few data mining algorithms produce, as part of their results, a measure of the degree of surprisingness of discovered knowledge. However, these measures can be computed in a post-processing phase, as a form of additional evaluation of the quality of discovered knowledge, complementing (rather than replacing) statistical measures of discovered knowledge accuracy. This papers presents a review of four measures of classification-rule surprisi...

In the last few years, the data mining community has proposed a number of objective rule interestingness measures to select the most interesting rules, out of a large set of discovered rules. However, it should be recalled that objective measures are just an estimate of the true degree of interestingness of a rule to the user, the so-called real human interest. The latter is inherently subjective. Hence, it is not clear how effective, in practice, objective measures are. More precisely, the central question investigated in this paper is: "how effective objective rule interestingness measures are, in the sense of being a good estimate of the true, subjective degree of interestingness of a rule to the user?" This question is investigated by extensive experiments with 11 objective rule interestingness measures across eight real-world data sets.

The immense volume of web usage data that exists on web servers can be mined to generate association rules that contain the information about website visitor interests, which can then be utilized for enhancing the website effectiveness, or increasing the profit in e-commerce applications. One of the problems in web usage association rule mining is the generation of too many rules with extremely high statistical interestingness measures, which are not necessarily interesting to the domain expert. In this paper, we propose a modified version of added value as an interestingness measure for particular groups of association rules consisting of web pages that occur as sub-items of the web site common menus. Our real life data experiments show that the modified added value outperforms added value, resembling the domain expert defined interestingness more closely.

The work aims to discover frequent patterns by generating the candidates and frame the association rules after which filter out only the efficient rules based on various Rule Interestingness measures. As all these require heavy computation, application of complete parallelization to every individual phase would yield better performance. The paper illustrates the system behavior in a heterogeneous environment with both shared memory and distributed memory parallelization while efficiently mining the data.

Association rules is a popular and well researched method for discovering interesting relation between variables in large databases and association rules, is one of the most important tasks in data mining. The generated strong association rules is depend on the association rule extraction by any algorithm, for example Apriory algorithm or Fp growth etc and the evolution of the rules by interestingness measure, for example support and confidence, lift or interest, correlation coefficient, statistical correlation, leverage, conviction etc. The association rules mining are dependent on both steps equally. The classical model of association rules mining is support-confidence, the interestingness measure of which is the confidence measure. The classical interestingness measure in Association Rules have existed some disadvantage. This paper present measurements that are support and confidence, interest or lift, chi-square test for independency, correlation coefficient, statistical correlation to calculate the strength of association rules. There are other interestingness measures, besides support and confidence which include generality, reliability, peculiarity, novelty, surprising ness, utility, and applicability. This paper investigates the evolution association rule mining.

Measures of interestingness play a crucial role in association rule mining. An important methodological problem, on which several papers appeared in the literature, is to provide a reasonable classication of the measures. In this paper, we explore Boolean factor analysis, which uses formal concepts corresponding to classes of measures as factors, for the purpose of clustering of the measures. Unlike the existing studies, our method reveals overlapping clusters of interestingness measures. We argue that the overlap between clusters is a desired feature of natural groupings of measures and that because formal concepts are used as factors in Boolean factor analysis, the resulting clusters have a clear meaning and are easy to interpret. We conduct three case studies on clustering of measures, provide interpretations of the resulting clusters and compare the results to those of the previous approaches reported in the literature.

Association rule mining of the web usage log files can be used to extract patterns of a website visito rs’ behavior. This knowledge can then be utilized to enhance web marketing strategies or improve the web browsing ex perience. In this paper we apply association rule mining on t he web usage log file of an educational institution . We use confidence and lift as the association rule interes tingness measures and compare their values in two d ifferent time periods. We show how this comparison brings additio nal information about association rules discovered and helps a webmaster make more informed decisions about the website enhancements.

The immense volume of web usage data that exists on web servers contains potentially valuable information about the behavior of website visitors. This information can be exploited in various ways, such as enhancing the effectiveness of websites or developing directed web marketing campaigns. In this paper we will focus on applying association rules as a data mining technique to extract potentially useful knowledge from web usage data. We conducted a comprehensive analysis of web usage association rules found on a website of an educational institution. Our experiments confirm that, prior to pruning, the set of generated association rules contained too many non-interesting rules, which made it very difficult for a user to find and exploit useful information. Many of these rules are a simple consequence of the high correlation between web pages due to their interconnectedness through the website link structure. We proposed and applied a set of basic pruning schemes to reduce the rule set size and to remove a significant number of non-interesting rules. This pruning method decreased the size of our experimental rule set by more than three times, making it much simpler to browse for truly interesting rules. The percentage of truly interesting rules, which can initiate a webmaster to actions that can potentially enhance the website and improve its browsing experience, in our resulting experimental rule set was 41%. The analysis of association rules in our case study confirmed the hypothesis that discovering interesting and potentially useful association rules in web usage data does not have to be a timeconsuming task and can lead to actions that increase the website's effectiveness.

Web usage log files generated on web servers contain huge amount of information that can be used for discovering web usage association rules, which can potentially give useful knowledge to the web usage data analysts. Association rule over-generation is a common problem in association rule mining that is further aggravated in web usage log mining due to the interconnectedness of web pages through the website link structure. We implemented a system for the discovery of association rules in web log usage data as an object-oriented application and used it to experiment on a real life web usage log data set. We proposed to alleviate the problem of web usage association rule over-generation by pruning the rules that contain directly linked pages out of the rule set. Our experiments showed that interestingness measures can successfully be used to sort the discovered association rules after the pruning method was applied. Most of the rules that ranked highly according to the interestingness measures proved to be truly valuable to a web master. We compared confidence and lift interestingness measures and found that lift outperformed confidence, but only after the minimum confidence threshold was taken into account.

Automatic discovery of web usage association rules is commonly used to extract the knowledge about web site visitors' interests. Its drawback is the generation of too many not truly interesting rules that have high statistical interestingness measures. We propose a method to prune rules that are statistically insignificant with respect to more general rules. Such rules may exist in the presence of high-confidence rules, which is often the case in web usage data. The method effectiveness is validated on two real-life web usage data sets.

Recently, a number of learning algorithms have been adapted for label ranking, including instance-based and tree-based methods. In this paper, we continue this line of work by proposing an adaptation of association rules for label ranking based on the APRIORI algorithm. Given that the original APRIORI algorithm does not aim to obtain predictive models, two changes were needed for this achievement. The adaptation essentially consists of using variations of the support and confidence measures based on ranking similarity functions that are suitable for label ranking. Additionally we propose a simple greedy method to select the parameters of the algorithm. We also adapt the method to make a prediction from the possibly conflicting consequents of the rules that apply to an example. Despite having made our adaptation from a very simple variant of association rules for classification, partial results clearly show that the method is making valid predictions. Additionally, they show that it competes well with state-of-the-art label ranking algorithms.

User preference is very important in orienting data miner, and this is the reason why these user preferences are integrated in the mining process, where they are coupled with Association Rules Mining "ARM" Algorithms to select only Association Rules "ARs" that satisfy the user's wishes and expectations. Within this framework, several approaches were proposed to overcome some problems which persist with the traditional ARM algorithms mainly dimensionality phenomenon engendered by thresholding and the subjective choice of measures. "MDP REF Algorithm" is one of these approaches; it prunes, filters to select the relevant ARs, while "Rank-Sort-MDP REF " sorts, ranks, and stores ARs to complete the MDP REF algorithm mining operation. Experiment result on real database showed the advantages of MDP REF algorithm and Rank-Sort-MDP REF algorithm over the other algorithms.

Using the association rules in datamining is one of the most relevant techniques in modern society, aiming to extract the interesting correlation and relation among sets of items or products in large transactional databases. The huge number of extracted association rules represents the main problem that a decision maker can face. Hence, the knowledge post-processing phase becomes very important and challenging to define the most interesting association rules, many interestingness measures have been proposed. Currently, there is no optimal measure that can be selected to evaluate the extracted association rules. To bypass this problem, we propose an approach based on multi-criteria optimization aiming to find a good compromise without excluding any measures. The experiments performed on numerous benchmark datasets show that the proposed algorithm is properly reducing a large number of association rules and keeping the most significant and interesting ones compared to other approaches which illustrate the efficiency and the applicability our approach.

La recherche de regles d’association interessantes est un domaine important et actif en fouille de donnees. Puisque les algorithmes utilises en extraction de connaissances a partir de donnees (ECD), ont tendance a generer un nombre important de regles, il est difficile a l’utilisateur de selectionner par lui meme les connaissances reellement interessantes. Pour repondre a ce probleme, un post-filtrage automatique des regles s’avere essentiel pour reduire fortement leur nombre. D’ou la proposition de nombreuses mesures d’interet dans la litterature, parmi lesquelles l’utilisateur est suppose choisir celle qui est la plus appropriee a ses objectifs. Comme l’interet depend a la fois des preferences de l’utilisateur et des donnees, les mesures ont ete repertoriees en deux categories : les mesures subjectives (orientees utilisateur ) et les mesures objectives (orientees donnees). Nous nous focalisons sur l’etude des mesures objectives. Neanmoins, il existe une plethore de mesures objecti...

With the rapid development of computer networks and information technology, an attacker has taken advantage to manipulate the situation to launch a complicated cyberattack. This complicated cyberattack causes a lot of problems among the organization because it requires an effective cyberattack attribution to mitigate and reduce the infection rate. Cyber Threat Intelligence (CTI) has gain wide coverage from the media due to its capability to provide CTI feeds from various data sources that can be used for cyberattack attribution. In this paper, we study the relationship of basic Indicator of Compromise (IOC) based on a network traffic dataset from a data mining approach. This dataset is obtained using a crawler that is deployed to pull security feed from Shadowserver. Then an association analysis method using Apriori Algorithm is implemented to extract rules that can discover interesting relationship between large sets of data items. Finally, the extracted rules are evaluated over the factor of interestingness measure of support, confidence and lift to quantify the value of association rules generated with Apriori Algorithm. By implementing the Apriori Algorithm in Shadowserver dataset, we discover some association rules among several IOC which can help attribute the cyberattack.

Fig. 1. Overview of a SOMFlow clustering graph that was created during our expert study to analyze speech intonation: First, a gender effect is identified (A) and removed using a domain-specific semitone normalization (B). The analyst created more detailed SOMs for artificial cells and added manual annotations (C) to filter noise caused by measurement errors. The resulting SOM reveals a relation to the pitch meta-attribute (D) and further data partitions allow the analyst to compare pitch contours of different speaker groups.

The assessment of the interestingness of sequential rules (generally temporal rules) is a crucial problem in sequence analysis. Due to their unsupervised nature, frequent pattern mining algorithms commonly generate a huge number of rules. However, while association rule interestingness has been widely studied in the literature, there are few measures dedicated to sequential rules. In this article, we propose an original statistical measure for assessing sequential rule interestingness. This measure named Sequential Implication Intensity (SII) evaluates the statistical significance of the rules in comparison with a probabilistic model. Numerical simulations show that SII has unique features for a sequential rule interestingness measure.

The assessment of the interestingness of sequential rules (generally temporal rules) is a crucial problem in sequence analysis. Due to their unsupervised nature, frequent pattern mining algorithms commonly generate a huge number of rules. However, while association rule interestingness has been widely studied in the literature, there are few measures dedicated to sequential rules. In this article, we propose an original statistical measure for assessing sequential rule interestingness. This measure named Sequential Implication Intensity (SII) evaluates the statistical significance of the rules in comparison with a probabilistic model. Numerical simulations show that SII has unique features for a sequential rule interestingness measure.

Web Usage Mining make use of Association Rule Mining to discover the interesting pattern, identify web user behavior, predict web user expectation and improve the business strategy. Association Rule Mining is a technique of Data Mining which is used to find the relationship between the data items. In Web Usage Mining, data are stored in the web server in the form of web log files. Numerous amounts of website visitors visit the web sites. So, it is not easy to access the web log files and find the relationship among them because of the rapid growth of web log files. Some preprocessing works are needed to reduce the noisy data of web log files before applying the association rules to find the relationship between the log files. Many researchers done the variety of works on web content mining and web usage mining to improve the efficiency of the websites by providing novel methods and this paper gives an overview about the existing works done by the researchers on web usage mining.

Fig. 1. Overview of a SOMFlow clustering graph that was created during our expert study to analyze speech intonation: First, a gender effect is identified (A) and removed using a domain-specific semitone normalization (B). The analyst created more detailed SOMs for artificial cells and added manual annotations (C) to filter noise caused by measurement errors. The resulting SOM reveals a relation to the pitch meta-attribute (D) and further data partitions allow the analyst to compare pitch contours of different speaker groups.

In this digital age, organizations have to deal with huge amounts of data, sometimes called Big Data. In recent years, the volume of data has increased substantially. Consequently, finding efficient and automated techniques for discovering useful patterns and relationships in the data becomes very important. In data mining, patterns and relationships can be represented in the form of association rules. Current techniques for discovering association rules rely on measures such as support for finding frequent patterns and confidence for finding association rules. A shortcoming of confidence is that it does not capture the correlation that exists between the left-hand side (LHS) and the right-hand side (RHS) of an association rule. On the other hand, the interestingness measure lift captures such as correlation in the sense that it tells us whether the LHS influences the RHS positively or negatively. Therefore, using Lift instead of confidence as a criteria for discovering association ...

In this digital age, organizations have to deal with huge amounts of data, sometimes called Big Data. In recent years, the volume of data has increased substantially. Consequently, finding efficient and automated techniques for discovering useful patterns and relationships in the data becomes very important. In data mining, patterns and relationships can be represented in the form of association rules. Current techniques for discovering association rules rely on measures such as support for finding frequent patterns and confidence for finding association rules. A shortcoming of confidence is that it does not capture the correlation that exists between the left-hand side (LHS) and the right-hand side (RHS) of an association rule. On the other hand, the interestingness measure lift captures such as correlation in the sense that it tells us whether the LHS influences the RHS positively or negatively. Therefore, using Lift instead of confidence as a criteria for discovering association rules can be more effective. It also gives the user more choices in determining the kind of association rules to be discovered. This in turn helps to narrow down the search space and consequently, improves performance. In this paper, we describe a new approach for discovering association rules that is based on Lift and not based on confidence.

Algorithmic probability is traditionally defined by considering the output of a universal machine fed with random programs. This definition proves inappropriate for many practical applications where probabilistic assessments are spontaneously and instantaneously performed. In particular, it does not tell what aspects of a situation are relevant when considering its probability ex-post (after its occurrence). As it stands, the standard definition also fails to capture the fact that simple, rather than complex outcomes are often considered improbable, as when a supposedly random device produces a repeated pattern. More generally, the standard algorithmic definition of probability conflicts with the idea that entropy maximum corresponds to states that are both complex (unordered) and probable. We suggest here that algorithmic probability should rather be defined as a difference in complexity. We distinguish description complexity from generation complexity. Improbable situations are si...

Mobile phones are pervasive, moderately specialized gadgets that have an effective and capable handling power enveloped with smaller segments that can do efficient and powerful calculations. One of the components that is built into the mobile phones to make it more robust are the sensors. Mobile phones are encompassed with several sensors, for example, proximity sensors, temperature sensors, accelerometers, gyroscopes and many more. These sensors have opened up ways to different fields in data mining and data analytics. The existence of these sensors has empowered people to control its information to perform different tasks. One such task is movement detection which is termed as activity recognition. In this paper an existing dataset has been used which consists of 10 volunteers, wearing a pair of accelerometers and gyroscopes close to their right lower arm and a pair of accelerometers and gyroscopes close to their left ankle. The subjects are asked to perform 12 exercises which are standing still, sitting and relaxing, lying down, walking, climbing stairs, waist-bends forward, frontal elevation of arms, knees bending (crouching), cycling, jogging, running, jumping front & back. 11 features were separated for the raw data collected from the sensors. In this paper, a novel automated method for classification of human activities, using wearable sensors which are also found interfaced within most of the modern mobile phones, is developed. The features are extracted from the recordings of data from individual as well as combination of sensors. The publicly available dataset is used for experimentation. The extracted features are classified using six popular classifiers: K-Nearest Neighbors (KNN), Naïve Bayes (NB), Support Vector Machine (SVM), Conditional Inference Tree (C-Tree), J48 and Random Forest (RF). The experimental results are tabulated and analyzed. Activity recognition turns out to be critical in distinguishing and sending fast data about irregular physical body developments of a person.

Assessing rules with interestingness measures is the cornerstone of successful applications of association rule discovery. However, there exists no information-theoretic measure which is adapted to the semantics of association rules. In this article, we present the Directed Information Ratio (DIR), a new rule interestingness measure which is based on information theory. DIR is specially designed for association rules, and in particular it differentiates two opposite rules a → b and a → b. Moreover, to our knowledge, DIR is the only rule interestingness measure which rejects both independence and (what we call) equilibrium, i.e. it discards both the rules whose antecedent and consequent are negatively correlated, and the rules which have more counterexamples than examples. Experimental studies show that DIR is a very filtering measure, which is useful for association rule post-processing.

The Association Rules Discovery is a technique widely used for various objectives. One is for Classification Based on Associations (CBA) with Class Association Rules (CARs). The number of rules discovered from data is extremely high with exponential numbers related to item types in the data. Thus, pruning uninteresting rules is a very important task with this technique. In the traditional technique, minimum Support and minimum Confidence are the main interestingness measures defined by the user for pruning tasks. However, some interesting rules have low Support or Confidence and are pruned at the same time as uninteresting rules. This problem usually occurs with an imbalanced Class ratio in a dataset such as the Scientific or Health dataset, positive-Class CARs usually have a much smaller number than negative-Class CARs. Positive-Class CARs are usually found to have low Support or Confidence that need trust in use without uninteresting rules. In this paper, we describe this problem in relation to a breast cancer dataset, and use a pruning task to discover interesting positive CARs even with low Support or Confidence. We propose a new measure called the Profitability-of-Interestingness Measure (PoI) to prune positive-class CARs from the dataset. Performance is measured by accuracy, precision, and recall. The results show that Pruned CARs have similar accuracy to traditional CARs. A comparison of the same rules for CBA Classifiers shows that Pruned CARs offer more precision than traditional CARs. The Pruned CARs set is more concise and easier to understand because of the lower number of confusing rules.

This paper presents a preprocessing step in mining association rules which uses tables to summarize synthetically the way variables interact by highlighting any zones which are attractive. Attractive zones are those which guarantee that potentially interesting rules will be extracted, and any irrelevant rules removed. These attractive zones will also make it possible to carry out a contextual discretization. In addition they constitute the starting point for mining association rules thereby decreasing the space where rules have to be searched for. Finally, this tabular representation of the behaviour of associations is particularly interesting in the case of quantitative variables where knowledge is no longer

This paper presents a preprocessing step in mining association rules which uses tables to summarize synthetically the way variables interact by highlighting any zones which are attractive. Attractive zones are those which guarantee that potentially interesting rules will be extracted, and any irrelevant rules removed. These attractive zones will also make it possible to carry out a contextual discretization. In addition they constitute the starting point for mining association rules thereby decreasing the space where rules have to be searched for. Finally, this tabular representation of the behaviour of associations is particularly interesting in the case of quantitative variables where knowledge is no longer parsed.

The development of good measures of interestingness of the discovered rules is one of the important problems in data mining. Such measures of interestingness are divided into objective measures :-those that depend only on the structure of a rule and the underlying data used in the discovery process, and the subjective measures-those that depend on the class of users who examine the rule. However, most objective measures are suitable for binary attributes and require an appropriate transformation of the initial set of attributes into binary attributes for all unsupervised usual algorithms for the discovery of association rules. As a result, the complexity of these algorithms increases exponentially with the number of attributes, and this transformation can lead us, on the one hand to a combinatorial explosion, and on the other hand to a prohibitive number of weakly significant rules with many redundancies. Moreover, the few measures suitable for numeric attributes, like for example correlation coefficient, are not selective. In this paper, we propose a new objective measure, called ordinal intensity of implication, which generalizes intensity of implication suitable for binary attributes and which evaluates whether the number of transactions not clearly verifying rule X→Y (i.e., the number of transactions containing a high value for attribute X and a low value for attribute Y) is significantly small as compared to a random draw. We finish the study with an evaluation on banking data and show some discovered ordinal rules, and connection between data / information and quality.

Web usage log files generated on web servers contain huge amount of information that can be used for discovering web usage association rules, which can potentially give useful knowledge to the web usage data analysts. Association rule over-generation is a common problem in association rule mining that is further aggravated in web usage log mining due to the interconnectedness of web pages through the website link structure. We implemented a system for the discovery of association rules in web log usage data as an object-oriented application and used it to experiment on a real life web usage log data set. We proposed to alleviate the problem of web usage association rule over-generation by pruning the rules that contain directly linked pages out of the rule set. Our experiments showed that interestingness measures can successfully be used to sort the discovered association rules after the pruning method was applied. Most of the rules that ranked highly according to the interestingness measures proved to be truly valuable to a web master. We compared confidence and lift interestingness measures and found that lift outperformed confidence, but only after the minimum confidence threshold was taken into account.

Data mining algorithms, especially those used for unsupervised learning, generate a large quantity of rules. In particular this applies to the Apriori family of algorithms for the determination of association rules. It is hence impossible for an expert in the field being mined to sustain these rules. To help carry out the task, many measures which evaluate the interestingness of rules have been developed. They make it possible to filter and sort automatically a set of rules with respect to given goals. Since these measures may produce different results, and as experts have different understandings of what a good rule is, we propose in this article a new direction to select the best rules: a two-step solution to the problem of the recommendation of one or more user-adapted interestingness measures. First, a description of interestingness measures, based on meaningful classical properties, is given. Second, a multicriteria decision aid process is applied to this analysis and illustrates the benefit that a user, who is not a data mining expert, can achieve with such methods.

Data mining is the efficient discovery of patterns in large databases, and classification rules are perhaps the most important type of patterns in data mining applications. However, the number of such classification rules is generally too huge that selection of interesting ones among all discovered rules becomes an important task. In this paper, factors related to the interestingness of a rule are investigated and some new factors are proposed. Following this, an interactive rule interestingness-learning algorithm (IRIL) is developed to automatically label the classification rules either as "interesting" or "uninteresting" with limited user participation. In our study, VFPI (Voting Feature Point Intervals), a feature projection based concept description learning and classification algorithm, is also developed in the framework of IRIL. Being specific to our concerns, VFPI takes the rule interestingness factors as features and is used to learn the rule interestingness concept and to classify the unlabeled classification rules. The success of the previously developed feature projection based learning and classification techniques encouraged us to develop VFPI. The empirical results based on TCMB (Central Bank of Turkish Republic) data set give promising results.