Bayesian network classifiers: Beyond classification accuracy

This work proposes and discusses an approach for inducing Bayesian classifiers aimed at balancing the tradeoff between the precise probability estimates produced by time consuming unrestricted Bayesian networks and the computational efficiency of Naive Bayes (NB) classifiers. The proposed approach is based on the fundamental principles of the Heuristic Search Bayesian network learning. The Markov Blanket concept, as well as a proposed ""approximate Markov Blanket"" are used to reduce the number of nodes that form the Bayesian network to be induced from data. Consequently, the usually high computational cost of the heuristic search learning algorithms can be lessened, while Bayesian network structures better than NB can be achieved. The resulting algorithms, called DMBC (Dynamic Markov Blanket Classifier) and A-DMBC (Approximate DMBC), are empirically assessed in twelve domains that illustrate scenarios of particular interest. The obtained results are compared with NB and Tree Augmented Network (TAN) classifiers, and confinn that both proposed algorithms can provide good classification accuracies and better probability estimates than NB and TAN, while being more computationally efficient than the widely used K2 Algorithm.

Detecting reliable gene interactions by a hierarchy of Bayesian network classifiers

The main purpose of a gene interaction network is to map the relationships of the genes that are out of sight when a genomic study is tackled. DNA microarrays allow the measure of gene expression of thousands of genes at the same time. These data constitute the numeric seed for the induction of the gene networks. In this paper, we propose a new approach to build gene networks by means of Bayesian classifiers, variable selection and bootstrap resampling. The interactions induced by the Bayesian classifiers are based both on the expression levels and on the phenotype information of the supervised variable. Feature selection and bootstrap resampling add reliability and robustness to the overall process removing the false positive findings. The consensus among all the induced models produces a hierarchy of dependences and, thus, of variables. Biologists can define the depth level of the model hierarchy so the set of interactions and genes involved can vary from a sparse to a dense set. Experimental results show how these networks perform well on classification tasks. The biological validation matches previous biological findings and opens new hypothesis for future studies

Using Bayesian networks with rule extraction to infer the risk of weed infestation in a corn-crop

This paper describes the modeling of a weed infestation risk inference system that implements a collaborative inference scheme based on rules extracted from two Bayesian network classifiers. The first Bayesian classifier infers a categorical variable value for the weed-crop competitiveness using as input categorical variables for the total density of weeds and corresponding proportions of narrow and broad-leaved weeds. The inferred categorical variable values for the weed-crop competitiveness along with three other categorical variables extracted from estimated maps for the weed seed production and weed coverage are then used as input for a second Bayesian network classifier to infer categorical variables values for the risk of infestation. Weed biomass and yield loss data samples are used to learn the probability relationship among the nodes of the first and second Bayesian classifiers in a supervised fashion, respectively. For comparison purposes, two types of Bayesian network structures are considered, namely an expert-based Bayesian classifier and a naive Bayes classifier. The inference system focused on the knowledge interpretation by translating a Bayesian classifier into a set of classification rules. The results obtained for the risk inference in a corn-crop field are presented and discussed. (C) 2009 Elsevier Ltd. All rights reserved.

Versatile Bayesian classifier for moving object detection by non-parametric background-foreground modeling

Along the recent years, several moving object detection strategies by non-parametric background-foreground modeling have been proposed. To combine both models and to obtain the probability of a pixel to belong to the foreground, these strategies make use of Bayesian classifiers. However, these classifiers do not allow to take advantage of additional prior information at different pixels. So, we propose a novel and efficient alternative Bayesian classifier that is suitable for this kind of strategies and that allows the use of whatever prior information. Additionally, we present an effective method to dynamically estimate prior probability from the result of a particle filter-based tracking strategy.

To select or to weigh: a comparative study of linear combination schemes for superparent-one-dependence estimators

We conduct a large-scale comparative study on linearly combining superparent-one-dependence estimators (SPODEs), a popular family of seminaive Bayesian classifiers. Altogether, 16 model selection and weighing schemes, 58 benchmark data sets, and various statistical tests are employed. This paper's main contributions are threefold. First, it formally presents each scheme's definition, rationale, and time complexity and hence can serve as a comprehensive reference for researchers interested in ensemble learning. Second, it offers bias-variance analysis for each scheme's classification error performance. Third, it identifies effective schemes that meet various needs in practice. This leads to accurate and fast classification algorithms which have an immediate and significant impact on real-world applications. Another important feature of our study is using a variety of statistical tests to evaluate multiple learning methods across multiple data sets.

Detection of severe obstructive sleep apnea through voice analysis

tThis paper deals with the potential and limitations of using voice and speech processing to detect Obstruc-tive Sleep Apnea (OSA). An extensive body of voice features has been extracted from patients whopresent various degrees of OSA as well as healthy controls. We analyse the utility of a reduced set offeatures for detecting OSA. We apply various feature selection and reduction schemes (statistical rank-ing, Genetic Algorithms, PCA, LDA) and compare various classifiers (Bayesian Classifiers, kNN, SupportVector Machines, neural networks, Adaboost). S-fold crossvalidation performed on 248 subjects showsthat in the extreme cases (that is, 127 controls and 121 patients with severe OSA) voice alone is able todiscriminate quite well between the presence and absence of OSA. However, this is not the case withmild OSA and healthy snoring patients where voice seems to play a secondary role. We found that thebest classification schemes are achieved using a Genetic Algorithm for feature selection/reduction.

Analyse de changements multiples : une approche probabiliste utilisant les réseaux bayésiens

La maintenance du logiciel est une phase très importante du cycle de vie de celui-ci. Après les phases de développement et de déploiement, c’est celle qui dure le plus longtemps et qui accapare la majorité des coûts de l'industrie. Ces coûts sont dus en grande partie à la difficulté d’effectuer des changements dans le logiciel ainsi que de contenir les effets de ces changements. Dans cette perspective, de nombreux travaux ont ciblé l’analyse/prédiction de l’impact des changements sur les logiciels. Les approches existantes nécessitent de nombreuses informations en entrée qui sont difficiles à obtenir. Dans ce mémoire, nous utilisons une approche probabiliste. Des classificateurs bayésiens sont entraînés avec des données historiques sur les changements. Ils considèrent les relations entre les éléments (entrées) et les dépendances entre changements historiques (sorties). Plus spécifiquement, un changement complexe est divisé en des changements élémentaires. Pour chaque type de changement élémentaire, nous créons un classificateur bayésien. Pour prédire l’impact d’un changement complexe décomposé en changements élémentaires, les décisions individuelles des classificateurs sont combinées selon diverses stratégies. Notre hypothèse de travail est que notre approche peut être utilisée selon deux scénarios. Dans le premier scénario, les données d’apprentissage sont extraites des anciennes versions du logiciel sur lequel nous voulons analyser l’impact de changements. Dans le second scénario, les données d’apprentissage proviennent d’autres logiciels. Ce second scénario est intéressant, car il permet d’appliquer notre approche à des logiciels qui ne disposent pas d’historiques de changements. Nous avons réussi à prédire correctement les impacts des changements élémentaires. Les résultats ont montré que l’utilisation des classificateurs conceptuels donne les meilleurs résultats. Pour ce qui est de la prédiction des changements complexes, les méthodes de combinaison "Voting" et OR sont préférables pour prédire l’impact quand le nombre de changements à analyser est grand. En revanche, quand ce nombre est limité, l’utilisation de la méthode Noisy-Or ou de sa version modifiée est recommandée.

Optimum-Path Forest Classifier for Large Scale Biometric Applications

This paper addresses biometric identification using large databases, in particular, iris databases. In such applications, it is critical to have low response time, while maintaining an acceptable recognition rate. Thus, the trade-off between speed and accuracy must be evaluated for processing and recognition parts of an identification system. In this paper, a graph-based framework for pattern recognition, called Optimum-Path Forest (OPF), is utilized as a classifier in a pre-developed iris recognition system. The aim of this paper is to verify the effectiveness of OPF in the field of iris recognition, and its performance for various scale iris databases. The existing Gauss-Laguerre Wavelet based coding scheme is used for iris encoding. The performance of the OPF and two other - Hamming and Bayesian - classifiers, is compared using small, medium, and large-scale databases. Such a comparison shows that the OPF has faster response for large-scale databases, thus performing better than the more accurate, but slower, classifiers.

Automatic microstructural characterization and classification using artificial intelligence techniques on ultrasound signals

Secondary phases such as Laves and carbides are formed during the final solidification stages of nickel based superalloy coatings deposited during the gas tungsten arc welding cold wire process. However, when aged at high temperatures, other phases can precipitate in the microstructure, like the γ″ and δ phases. This work presents a new application and evaluation of artificial intelligent techniques to classify (the background echo and backscattered) ultrasound signals in order to characterize the microstructure of a Ni-based alloy thermally aged at 650 and 950 °C for 10, 100 and 200 h. The background echo and backscattered ultrasound signals were acquired using transducers with frequencies of 4 and 5 MHz. Thus with the use of features extraction techniques, i.e.; detrended fluctuation analysis and the Hurst method, the accuracy and speed in the classification of the secondary phases from ultrasound signals could be studied. The classifiers under study were the recent optimum-path forest (OPF) and the more traditional support vector machines and Bayesian. The experimental results revealed that the OPF classifier was the fastest and most reliable. In addition, the OPF classifier revealed to be a valid and adequate tool for microstructure characterization through ultrasound signals classification due to its speed, sensitivity, accuracy and reliability. © 2013 Elsevier B.V. All rights reserved.

EEG signal classification for epilepsy diagnosis via optimum path forest - A systematic assessment

Conselho Nacional de Desenvolvimento Científico e Tecnológico (CNPq)

Machine Learning applicato al Web Semantico: Statistical Relational Learning vs Tensor Factorization

Obiettivo della tesi è analizzare e testare i principali approcci di Machine Learning applicabili in contesti semantici, partendo da algoritmi di Statistical Relational Learning, quali Relational Probability Trees, Relational Bayesian Classifiers e Relational Dependency Networks, per poi passare ad approcci basati su fattorizzazione tensori, in particolare CANDECOMP/PARAFAC, Tucker e RESCAL.

Identification of a biomarker panel for colorectal cancer diagnosis

Background Malignancies arising in the large bowel cause the second largest number of deaths from cancer in the Western World. Despite progresses made during the last decades, colorectal cancer remains one of the most frequent and deadly neoplasias in the western countries. Methods A genomic study of human colorectal cancer has been carried out on a total of 31 tumoral samples, corresponding to different stages of the disease, and 33 non-tumoral samples. The study was carried out by hybridisation of the tumour samples against a reference pool of non-tumoral samples using Agilent Human 1A 60-mer oligo microarrays. The results obtained were validated by qRT-PCR. In the subsequent bioinformatics analysis, gene networks by means of Bayesian classifiers, variable selection and bootstrap resampling were built. The consensus among all the induced models produced a hierarchy of dependences and, thus, of variables. Results After an exhaustive process of pre-processing to ensure data quality--lost values imputation, probes quality, data smoothing and intraclass variability filtering--the final dataset comprised a total of 8, 104 probes. Next, a supervised classification approach and data analysis was carried out to obtain the most relevant genes. Two of them are directly involved in cancer progression and in particular in colorectal cancer. Finally, a supervised classifier was induced to classify new unseen samples. Conclusions We have developed a tentative model for the diagnosis of colorectal cancer based on a biomarker panel. Our results indicate that the gene profile described herein can discriminate between non-cancerous and cancerous samples with 94.45% accuracy using different supervised classifiers (AUC values in the range of 0.997 and 0.955)

Identification of a biomarker panel for colorectal cancer diagnosis

Background:Malignancies arising in the large bowel cause the second largest number of deaths from cancer in the Western World. Despite progresses made during the last decades, colorectal cancer remains one of the most frequent and deadly neoplasias in the western countries. Methods: A genomic study of human colorectal cancer has been carried out on a total of 31 tumoral samples, corresponding to different stages of the disease, and 33 non-tumoral samples. The study was carried out by hybridisation of the tumour samples against a reference pool of non-tumoral samples using Agilent Human 1A 60-mer oligo microarrays. The results obtained were validated by qRT-PCR. In the subsequent bioinformatics analysis, gene networks by means of Bayesian classifiers, variable selection and bootstrap resampling were built. The consensus among all the induced models produced a hierarchy of dependences and, thus, of variables. Results: After an exhaustive process of pre-processing to ensure data quality--lost values imputation, probes quality, data smoothing and intraclass variability filtering--the final dataset comprised a total of 8, 104 probes. Next, a supervised classification approach and data analysis was carried out to obtain the most relevant genes. Two of them are directly involved in cancer progression and in particular in colorectal cancer. Finally, a supervised classifier was induced to classify new unseen samples. Conclusions: We have developed a tentative model for the diagnosis of colorectal cancer based on a biomarker panel. Our results indicate that the gene profile described herein can discriminate between non-cancerous and cancerous samples with 94.45% accuracy using different supervised classifiers (AUC values in the range of 0.997 and 0.955).

The n-tuple network: coping with sub-optimality

The subject of this thesis is the n-tuple net.work (RAMnet). The major advantage of RAMnets is their speed and the simplicity with which they can be implemented in parallel hardware. On the other hand, this method is not a universal approximator and the training procedure does not involve the minimisation of a cost function. Hence RAMnets are potentially sub-optimal. It is important to understand the source of this sub-optimality and to develop the analytical tools that allow us to quantify the generalisation cost of using this model for any given data. We view RAMnets as classifiers and function approximators and try to determine how critical their lack of' universality and optimality is. In order to understand better the inherent. restrictions of the model, we review RAMnets showing their relationship to a number of well established general models such as: Associative Memories, Kamerva's Sparse Distributed Memory, Radial Basis Functions, General Regression Networks and Bayesian Classifiers. We then benchmark binary RAMnet. model against 23 other algorithms using real-world data from the StatLog Project. This large scale experimental study indicates that RAMnets are often capable of delivering results which are competitive with those obtained by more sophisticated, computationally expensive rnodels. The Frequency Weighted version is also benchmarked and shown to perform worse than the binary RAMnet for large values of the tuple size n. We demonstrate that the main issues in the Frequency Weighted RAMnets is adequate probability estimation and propose Good-Turing estimates in place of the more commonly used :Maximum Likelihood estimates. Having established the viability of the method numerically, we focus on providillg an analytical framework that allows us to quantify the generalisation cost of RAMnets for a given datasetL. For the classification network we provide a semi-quantitative argument which is based on the notion of Tuple distance. It gives a good indication of whether the network will fail for the given data. A rigorous Bayesian framework with Gaussian process prior assumptions is given for the regression n-tuple net. We show how to calculate the generalisation cost of this net and verify the results numerically for one dimensional noisy interpolation problems. We conclude that the n-tuple method of classification based on memorisation of random features can be a powerful alternative to slower cost driven models. The speed of the method is at the expense of its optimality. RAMnets will fail for certain datasets but the cases when they do so are relatively easy to determine with the analytical tools we provide.