In today’s digital era, establishing an online presence and maintaining a well-structured website is vitalfor companies to remain competitive in their respective markets. A crucial aspect of online success liesin strategically selecting the right words to optimize customer engagement and search engine visibility.However, this process is often time-consuming, involving extensive analysis of a company’s website aswell as its competitors’. This thesis focuses on developing an efficient binary classification approachto identify key words and phrases extracted from multiple company websites. The data set used forthis solution consists of approximately 92,000 scraped samples, primarily comprising non-key samples.Various features were extracted, and a word embedding model was employed to assess each sample’srelevance to its specific industry and topic. The logistic regression, decision tree and random forestalgorithms were all explored and implemented as different solutions to the classification problem. Theresults indicated that the logistic regression model excelled in retaining keywords but was less effectivein eliminating non-keywords. Conversely, the tree-based methods demonstrated superior classificationof keywords, albeit at the cost of misclassifying a few keywords. Overall, the random forest approachoutperformed the others, achieving a result of 76 percent in recall and 20 percent in precision whenpredicting key samples. In summary, this thesis presents a solution for classifying words and phrasesfrom company websites into key and non-key categories, and the developed methodology could offervaluable insights for companies seeking to enhance their website optimization strategies.

abstract: In many biological research studies, including speech analysis, clinical research, and prediction studies, the validity of the study is dependent on the effectiveness of the training data set to represent the target population. For example, in speech analysis, if one is performing emotion classification based on speech, the performance of the classifier is mainly dependent on the number and quality of the training data set. For small sample sizes and unbalanced data, classifiers developed in this context may be focusing on the differences in the training data set rather than emotion (e.g., focusing on gender, age, and dialect). This thesis evaluates several sampling methods and a non-parametric approach to sample sizes required to minimize the effect of these nuisance variables on classification performance. This work specifically focused on speech analysis applications, and hence the work was done with speech features like Mel-Frequency Cepstral Coefficients (MFCC) and Filter Bank Cepstral Coefficients (FBCC). The non-parametric divergence (D_p divergence) measure was used to study the difference between different sampling schemes (Stratified and Multistage sampling) and the changes due to the sentence types in the sampling set for the process. / Dissertation/Thesis / Masters Thesis Electrical Engineering 2020

Credit card fraud is a field with perpetrators performing illegal actions that may affect other individuals or companies negatively. For instance, a criminalcan steal credit card information from an account holder and then conduct fraudulent transactions. The activities are a potential contributory factor to how illegal organizations such as terrorists and drug traffickers support themselves financially. Within the machine learning area, there are several methods that possess the ability to detect credit card fraud transactions; supervised learning and unsupervised learning algorithms. This essay investigates the supervised approach, where two algorithms (Hellinger Distance Decision Tree (HDDT) and Random Forest) are evaluated on a real life dataset of 284,807 transactions. Under those circumstances, the main purpose is to develop a “well-functioning” model with a reasonable capacity to categorize transactions as fraudulent or legit. As the data is heavily unbalanced, reducing the false-positive rate is also an important part when conducting research in the chosen area. In conclusion, evaluated algorithms present a fairly similar outcome, where both models have the capability to distinguish the classes from each other. However, the Random Forest approach has a better performance than HDDT in all measures of interest.

As técnicas de mineração de dados, e mais especificamente de aprendizado de máquina, têm se popularizado enormemente nos últimos anos, passando a incorporar os Sistemas de Informação para Apoio à Decisão, Previsão de Eventos e Análise de Dados. Por exemplo, sistemas de apoio à decisão na área médica e ambientes de \\textit{Business Intelligence} fazem uso intensivo dessas técnicas. Algoritmos indutores de árvores de classificação, particularmente os algoritmos TDIDT (Top-Down Induction of Decision Trees), figuram entre as técnicas mais comuns de aprendizado supervisionado. Uma das vantagens desses algoritmos em relação a outros é que, uma vez construída e validada, a árvore tende a ser interpretada com relativa facilidade, sem a necessidade de conhecimento prévio sobre o algoritmo de construção. Todavia, são comuns problemas de classificação em que as frequências relativas das classes variam significativamente. Algoritmos baseados em minimização do erro global de classificação tendem a construir classificadores com baixas taxas de erro de classificação nas classes majoritárias e altas taxas de erro nas classes minoritárias. Esse fenômeno pode ser crítico quando as classes minoritárias representam eventos como a presença de uma doença grave (em um problema de diagnóstico médico) ou a inadimplência em um crédito concedido (em um problema de análise de crédito). Para tratar esse problema, diversos algoritmos TDIDT demandam a calibração de parâmetros {\\em ad-hoc} ou, na ausência de tais parâmetros, a adoção de métodos de balanceamento dos dados. As duas abordagens não apenas introduzem uma maior complexidade no uso das ferramentas de mineração de dados para usuários menos experientes, como também nem sempre estão disponíveis. Neste trabalho, propomos um novo algoritmo indutor de árvores de classificação para problemas com dados desbalanceados. Esse algoritmo, denominado atualmente DDBT (Dynamic Discriminant Bounds Tree), utiliza um critério de partição de nós que, ao invés de se basear em frequências absolutas de classes, compara as proporções das classes nos nós com as proporções do conjunto de treinamento original, buscando formar subconjuntos com maior discriminação de classes em relação ao conjunto de dados original. Para a rotulação de nós terminais, o algoritmo atribui a classe com maior prevalência relativa no nó em relação à prevalência no conjunto original. Essas características fornecem ao algoritmo a flexibilidade para o tratamento de conjuntos de dados com desbalanceamento de classes, resultando em um maior equilíbrio entre as taxas de erro em classificação de objetos entre as classes. / Data mining techniques and, particularly, machine learning methods, have become very popular in recent years. Many decision support information systems and business intelligence tools have incorporated and made intensive use of such techniques. Top-Down Induction of Decision Trees Algorithms (TDIDT) appear among the most popular tools for supervised learning. One of their advantages with respect to other methods is that a decision tree is frequently easy to be interpreted by the domain specialist, precluding the necessity of previous knowledge about the induction algorithms. On the other hand, several typical classification problems involve unbalanced data (heterogeneous class prevalence). In such cases, algorithms based on global error minimization tend to induce classifiers with low error rates over the high prevalence classes, but with high error rates on the low prevalence classes. This phenomenon may be critical when low prevalence classes represent rare or important events, like the presence of a severe disease or the default in a loan. In order to address this problem, several TDIDT algorithms require the calibration of {\\em ad-hoc} parameters, or even data balancing techniques. These approaches usually make data mining tools more complex for less expert users, if they are ever available. In this work, we propose a new TDIDT algorithm for problems involving unbalanced data. This algorithm, currently named DDBT (Dynamic Discriminant Bounds Tree), uses a node partition criterion which is not based on absolute class frequencies, but compares the prevalence of each class in the current node with those in the original training sample. For terminal nodes labeling, the algorithm assigns the class with maximum ration between the relative prevalence in the node and the original prevalence in the training sample. Such characteristics provide more flexibility for the treatment of unbalanced data-sets, yielding a higher equilibrium among the error rates in the classes.

As técnicas de mineração de dados, e mais especificamente de aprendizado de máquina, têm se popularizado enormemente nos últimos anos, passando a incorporar os Sistemas de Informação para Apoio à Decisão, Previsão de Eventos e Análise de Dados. Por exemplo, sistemas de apoio à decisão na área médica e ambientes de \\textit{Business Intelligence} fazem uso intensivo dessas técnicas. Algoritmos indutores de árvores de classificação, particularmente os algoritmos TDIDT (Top-Down Induction of Decision Trees), figuram entre as técnicas mais comuns de aprendizado supervisionado. Uma das vantagens desses algoritmos em relação a outros é que, uma vez construída e validada, a árvore tende a ser interpretada com relativa facilidade, sem a necessidade de conhecimento prévio sobre o algoritmo de construção. Todavia, são comuns problemas de classificação em que as frequências relativas das classes variam significativamente. Algoritmos baseados em minimização do erro global de classificação tendem a construir classificadores com baixas taxas de erro de classificação nas classes majoritárias e altas taxas de erro nas classes minoritárias. Esse fenômeno pode ser crítico quando as classes minoritárias representam eventos como a presença de uma doença grave (em um problema de diagnóstico médico) ou a inadimplência em um crédito concedido (em um problema de análise de crédito). Para tratar esse problema, diversos algoritmos TDIDT demandam a calibração de parâmetros {\\em ad-hoc} ou, na ausência de tais parâmetros, a adoção de métodos de balanceamento dos dados. As duas abordagens não apenas introduzem uma maior complexidade no uso das ferramentas de mineração de dados para usuários menos experientes, como também nem sempre estão disponíveis. Neste trabalho, propomos um novo algoritmo indutor de árvores de classificação para problemas com dados desbalanceados. Esse algoritmo, denominado atualmente DDBT (Dynamic Discriminant Bounds Tree), utiliza um critério de partição de nós que, ao invés de se basear em frequências absolutas de classes, compara as proporções das classes nos nós com as proporções do conjunto de treinamento original, buscando formar subconjuntos com maior discriminação de classes em relação ao conjunto de dados original. Para a rotulação de nós terminais, o algoritmo atribui a classe com maior prevalência relativa no nó em relação à prevalência no conjunto original. Essas características fornecem ao algoritmo a flexibilidade para o tratamento de conjuntos de dados com desbalanceamento de classes, resultando em um maior equilíbrio entre as taxas de erro em classificação de objetos entre as classes. / Data mining techniques and, particularly, machine learning methods, have become very popular in recent years. Many decision support information systems and business intelligence tools have incorporated and made intensive use of such techniques. Top-Down Induction of Decision Trees Algorithms (TDIDT) appear among the most popular tools for supervised learning. One of their advantages with respect to other methods is that a decision tree is frequently easy to be interpreted by the domain specialist, precluding the necessity of previous knowledge about the induction algorithms. On the other hand, several typical classification problems involve unbalanced data (heterogeneous class prevalence). In such cases, algorithms based on global error minimization tend to induce classifiers with low error rates over the high prevalence classes, but with high error rates on the low prevalence classes. This phenomenon may be critical when low prevalence classes represent rare or important events, like the presence of a severe disease or the default in a loan. In order to address this problem, several TDIDT algorithms require the calibration of {\\em ad-hoc} parameters, or even data balancing techniques. These approaches usually make data mining tools more complex for less expert users, if they are ever available. In this work, we propose a new TDIDT algorithm for problems involving unbalanced data. This algorithm, currently named DDBT (Dynamic Discriminant Bounds Tree), uses a node partition criterion which is not based on absolute class frequencies, but compares the prevalence of each class in the current node with those in the original training sample. For terminal nodes labeling, the algorithm assigns the class with maximum ration between the relative prevalence in the node and the original prevalence in the training sample. Such characteristics provide more flexibility for the treatment of unbalanced data-sets, yielding a higher equilibrium among the error rates in the classes.

No description available.

Les documents publiés par des entreprises, tels les communiqués de presse, contiennent une foule d’informations sur diverses activités des entreprises. C’est une source précieuse pour des analyses en intelligence d’affaire. Cependant, il est nécessaire de développer des outils pour permettre d’exploiter cette source automatiquement, étant donné son grand volume. Ce mémoire décrit un travail qui s’inscrit dans un volet d’intelligence d’affaire, à savoir la détection de relations d’affaire entre les entreprises décrites dans des communiqués de presse. Dans ce mémoire, nous proposons une approche basée sur la classification. Les méthodes de classifications existantes ne nous permettent pas d’obtenir une performance satisfaisante. Ceci est notamment dû à deux problèmes : la représentation du texte par tous les mots, qui n’aide pas nécessairement à spécifier une relation d’affaire, et le déséquilibre entre les classes. Pour traiter le premier problème, nous proposons une approche de représentation basée sur des mots pivots c’est-à-dire les noms d’entreprises concernées, afin de mieux cerner des mots susceptibles de les décrire. Pour le deuxième problème, nous proposons une classification à deux étapes. Cette méthode s’avère plus appropriée que les méthodes traditionnelles de ré-échantillonnage. Nous avons testé nos approches sur une collection de communiqués de presse dans le domaine automobile. Nos expérimentations montrent que les approches proposées peuvent améliorer la performance de classification. Notamment, la représentation du document basée sur les mots pivots nous permet de mieux centrer sur les mots utiles pour la détection de relations d’affaire. La classification en deux étapes apporte une solution efficace au problème de déséquilibre entre les classes. Ce travail montre que la détection automatique des relations d’affaire est une tâche faisable. Le résultat de cette détection pourrait être utilisé dans une analyse d’intelligence d’affaire. / Documents published by companies such as press releases, contain a wealth of information on various business activities. This is a valuable source for business intelligence analysis; but automatic tools are needed to exploit such large volume data. The work described in this thesis is part of a research project on business intelligence, namely we aim at the detection of business relationships between companies described in press releases. In this thesis, we consider business relation detection as a problem of classification. However, the existing classification methods do not allow us to obtain a satisfactory performance. This is mainly due to two problems: the representation of text using all the content words, which do not necessarily a business relationship; and the imbalance between classes. To address the first problem, we propose representations based on words that are between or close to the names of companies involved (which we call pivot words) in order to focus on words having a higher chance to describe a relation. For the second problem, we propose a two-stage classification. This method is more effective than the traditional resampling methods. We tested our approach on a collection of press releases in the automotive industry. Our experiments show that both proposed approaches can improve the classification performance. They perform much better than the traditional feature selection methods and the resampling method. This work shows the feasibility of automatic detection of business relations. The result of this detection could be used in an analysis of business intelligence.

<p dir="ltr">The parallel growth of contemporary machine learning (ML) technologies alongside edge/-fog networking has necessitated the development of novel paradigms to effectively manage their intersection. Specifically, the proliferation of edge devices equipped with data generation and ML model training capabilities has given rise to an alternative paradigm called federated learning (FL), moving away from traditional centralized ML common in cloud-based networks. FL involves training ML models directly on edge devices where data are generated.</p><p dir="ltr">A fundamental challenge of FL lies in the extensive heterogeneity inherent to edge/fog networks, which manifests in various forms such as (i) statistical heterogeneity: edge devices have distinct underlying data distributions, (ii) structural heterogeneity: edge devices have diverse physical hardware, (iii) data quality heterogeneity: edge devices have varying ratios of labeled and unlabeled data, and (iv) adversarial compromise: some edge devices may be compromised by adversarial attacks. This dissertation endeavors to capture and model these intricate relationships at the intersection of FL and highly heterogeneous edge/fog networks. To do so, this dissertation will initially develop closed-form expressions for the trade-offs between ML performance and resource cost considerations within edge/fog networks. Subsequently, it optimizes the fundamental processes of FL, encompassing aspects such as batch size control for stochastic gradient descent (SGD) and sampling for global aggregations. This optimization is jointly formulated with networking considerations, which include communication resource consumption and device-to-device (D2D) cooperation.</p><p dir="ltr">In the former half of the dissertation, the emphasis is first on optimizing device sampling for global aggregations in FL, and then on developing a self-sufficient hierarchical meta-learning approach for FL. These methodologies maximize expected ML model performance while addressing common challenges associated with statistical and system heterogeneity. Novel techniques, such as management of D2D data offloading, adaptive CPU clock cycle control, integration of meta-learning, and much more, enable these methodologies. In particular, the proposed hierarchical meta-learning approach enables rapid integration of new devices in large-scale edge/fog networks.</p><p dir="ltr">The latter half of the dissertation directs its ocus towards emerging forms of heterogeneity in FL scenarios, namely (i) heterogeneity in quantity and quality of local labeled and unlabeled data at edge devices and (ii) heterogeneity in terms of adversarially comprised edge devices. To deal with heterogeneous labeled/unlabeled data across edge networks, this dissertation proposes a novel methodology that enables multi-source to multi-target federated domain adaptation. This proposed methodology views edge devices as sources – devices with mostly labeled data that perform ML model training, or targets - devices with mostly unlabeled data that rely on sources’ ML models, and subsequently optimizes the network relationships. In the final chapter, a novel methodology to improve FL robustness is developed in part by viewing adversarial attacks on FL as a form of heterogeneity.</p>

L'objectif de cette thèse est de proposer une méthodologie statistique permettant de formuler une règle de classement capable de surmonter les difficultés qui se présentent dans le traitement des données lorsque la distribution a priori de la variable réponse est déséquilibrée. Notre proposition est construite autour d'un ensemble particulier de règles d'association appelées "class association rules". Dans le chapitre II, nous avons exposé les bases théoriques qui sous-tendent la méthode. Nous avons utilisé les indicateurs de performance usuels existant dans la littérature pour évaluer un classifieur. A chaque règle "class association rule" est associée un classifieur faible engendré par l'antécédent de la règle que nous appelons profils. L'idée de la méthode est alors de combiner un nombre réduit de classifieurs faibles pour constituer une règle de classement performante. Dans le chapitre III, nous avons développé les différentes étapes de la procédure d'apprentissage statistique lorsque les observations sont indépendantes et identiquement distribuées. On distingue trois grandes étapes: (1) une étape de génération d'un ensemble initial de profils, (2) une étape d'élagage de profils redondants et (3) une étape de sélection d'un ensemble optimal de profils. Pour la première étape, nous avons utilisé l'algorithme "apriori" reconnu comme l'un des algorithmes de base pour l'exploration des règles d'association. Pour la deuxième étape, nous avons proposé un test stochastique. Et pour la dernière étape un test asymptotique est effectué sur le rapport des valeurs prédictives positives des classifieurs lorsque les profils générateurs respectifs sont emboîtés. Il en résulte un ensemble réduit et optimal de profils dont la combinaison produit une règle de classement performante. Dans le chapitre IV, nous avons proposé une extension de la méthode d'apprentissage statistique lorsque les observations ne sont pas identiquement distribuées. Il s'agit précisément d'adapter la procédure de sélection de l'ensemble optimal lorsque les données ne sont pas identiquement distribuées. L'idée générale consiste à faire une estimation bayésienne de toutes les valeurs prédictives positives des classifieurs faibles. Par la suite, à l'aide du facteur de Bayes, on effectue un test d'hypothèse sur le rapport des valeurs prédictives positives lorsque les profils sont emboîtés. Dans le chapitre V, nous avons appliqué la méthodologie mise en place dans les chapitres précédents aux données du projet QUARITE concernant la mortalité maternelle au Sénégal et au Mali. / The aim of this thesis is to design a supervised statistical learning methodology that can overcome the weakness of standard methods when the prior distribution of the response variable is unbalanced. The proposed methodology is built using class association rules. Chapter II deals with theorical basis of statistical learning method by relating various classifiers performance metrics with class association rules. Since the classifier corresponding to a class association rules is a weak classifer, we propose to select a small number of such weak classifiers and to combine them in the aim to build an efficient classifier. In Chapter III, we develop the different steps of the statistical learning method when observations are independent and identically distributed. There are three main steps: In the first step, an initial set of patterns correlated with the target class is generated using "apriori" algorithm. In the second step, we propose a hypothesis test to prune redondant patterns. In the third step, an hypothesis test is performed based on the ratio of the positive predictive values of the classifiers when respective generating patterns are nested. This results in a reduced and optimal set of patterns whose combination provides an efficient classifier. In Chapter IV, we extend the classification method that we proposed in Chapter III in order to handle the case where observations are not identically distributed. The aim being here to adapt the procedure for selecting the optimal set of patterns when data are grouped data. In this setting we compute the estimation of the positive predictive values as the mean of the posterior distribution of the target class probability by using empirical Bayes method. Thereafter, using Bayes factor, a hypothesis test based on the ratio of the positive predictive values is carried out when patterns are nested. Chapter V is devoted to the application of the proposed methodology to process a real world dataset. We studied the QUARITE project dataset on maternal mortality in Senegal and Mali in order to provide a decision making tree that health care professionals can refer to when managing patients delivering in their health facilities.

This thesis focuses on problematics of segmentation of unbalanced datasets by the useof artificial inteligence. Numerous existing methods for dealing with unbalanced datasetsare examined, and some of them are then applied to real problem that consist of seg-mentation of dataset with class ratio of more than 6000:1.