Feature Selection

Face expression recognition technology is one of the most recently developed fields in machine learning and has profoundly helped its users through forensic, security, and biometric applications. Many researchers and program developers have allocated their time and energy to figure out various techniques which would add to the technology’s functionality and accuracy. Face expression recognition is a complicated computational process in which is implemented via analyzing changes in facial traits that follow different emotional reactions. This paper endeavors to inspect accuracy ratio of six classifiers based on Relief-F feature selection method, relying on the utilization of the minimum quantity of attributes. The classifiers in which the paper attempts to inspect are Multi-Layer Perceptron, Random Forest, Decision Tree, Support Vector Machine, K-Nearest Neighbor, and Radial Basis Function. The experiment illustrates that K-Nearest Neighbor is the most accurate classifier with the to...

In our previous study we have shown that identification of bacteria species with the use of Fresnel diffraction patterns is possible with high accuracy and at low cost. Fresnel diffraction patterns were recorded with the optical system with converging spherical wave illumination. Obtained experimental results have shown that colonies of specific bacteria species generate unique diffraction signatures. Features used for building classification models and thus for identification were simply mean value and standard deviation calculated of pixel intensities within regions of interest called rings. This work presents new, interpretable features denoting morphological and textural properties of the Fresnel diffraction patterns and their verification with the use of the statistical analysis workflow specially developed for bacteria species identification. As data set of bacteria species diffraction patterns it is very important to find features that differentiate species in the best manner. This task includes two steps. The first is finding and extracting new, interpretable features that can potentially be better for bacteria species differentiation than the ones used before. While the second one is deciding which of them are the best for identification purposes. The new features are calculated basing on normalized diffraction patterns and central statistical moments. For the verification the analysis workflow based on ANOVA for feature selection, LDA, QDA and SVM models for classification and identification and CV, sensitivity and specificity for performance assessment of the identification process, are applied. Additionally, the Fisher divergence method also known as signal to noise ratio (SNR) for feature selection was exploited.

Electronic Commerce is one of the most significant fields in internet appli- cations. With the focus moving from B2C-commerce to B2B-commerce, in- tegrating internet and legacy systems within one company and getting technologies used in different companies to work together is a pretentious task. After discussing the current research in software architecture, includ- ing suitable notations for describing them, the

Reuters Corpus Volume I (RCV1) is an archive of over 800,000 manually categorized newswire stories recently made available by Reuters, Ltd. for research purposes. Use of this data for research on text categorization requires a detailed understanding of the real world constraints under which the data was produced. Drawing on interviews with Reuters personnel and access to Reuters documentation, we describe the coding policy and quality control procedures used in producing the RCV1 data, the intended semantics of the hierarchical category taxonomies, and the corrections necessary to remove errorful data. We refer to the original data as RCV1-v1, and the corrected data as RCV1-v2. We benchmark several widely used supervised learning methods on RCV1-v2, illustrating the collection's properties, suggesting new directions for research, and providing baseline results for future studies. We make available detailed, per-category experimental results, as well as corrected versions of the category assignments and taxonomy structures, via online appendices.

Each complex network (or class of networks) presents specific topological features which characterize its connectivity and highly influence the dynamics of processes executed on the network. The analysis, discrimination, and synthesis of complex networks therefore rely on the use of measurements capable of expressing the most relevant topological features. This article presents a survey of such measurements. It includes general considerations about complex network characterization, a brief review of the principal models, and the ...

Correct prediction of the translation initiation site (TIS) is an important issue in genomic research. We show that feature generation together with correlation based feature selection can be used with a variety of machine learning algorithms to give highly accurate translation initiation site prediction. Only very few features are needed and the results achieve comparable accuracy to the best existing approaches. Our approach has the advantage that it does not require one to devise a special prediction method; rather standard machine learning classifiers are shown to give very good performance on the selected features. The raw and generated features which we have found to be important are the following: positions −3 and −1 in the sequence; upstream k-grams for k=3, 4, and 5; stop-codon frequency; downstream in-frame 3-gram; and the distance of ATG to the beginning of the sequence. The best result, with an overall accuracy of 90%, is obtained by selecting only seven features from this set. The same features retrained with the use of a scanning model achieves an overall accuracy of 94% on this dataset.

We propose a supervised approach to detect falls in home environment using an optimised descriptor adapted to real-time tasks. We introduce a realistic dataset of 191 videos, a new metric allowing to evaluate fall detection performance in a video stream, and an automatically optimised set of spatio-temporal descriptors which fed a supervised classifier. We build the initial spatio-temporal descriptor named ST HF using several combinations of transformations of geometrical features (height and width of human body bounding box, the user's trajectory with her/his orientation, projection histograms and moments of order 0, 1 and 2). We study the combinations of usual transformations of the features (Fourier Transform, Wavelet transform, first and second derivatives), and we show experimentally that it is possible to achieve high performance using Support Vector Machine and Adaboost classifiers. Automatic feature selection allows to show that the best tradeoff between classification performance and processing time is obtained combining the original low-level features with their first derivative. Hence, we evaluate the robustness of the fall detection regarding location changes. We propose a realistic and pragmatic protocol which enables performance to be improved by updating the training in the current location with normal activities records.

Increase in the amount of information on the Web has caused the need for accurate automated classifiers for Web pages to maintain Web directories and to increase search engines’ performance. As every (HTML/XML) tag and every term on each Web page can be considered as a feature, we need efficient methods to select best features to reduce feature space of the Web page classification problem. In this study, our aim is to apply a recent optimization technique namely the Intelligent Water Drops (IWD) algorithm, to select best features for Web page classification problem. IWD is a swarm based optimization algorithm which has been inspired from natural water drops in rivers. In this study, we use IWD to select a subset of features, and to evaluate the fitness of the selected features J48 classifier of the Weka data mining tool is employed. WebKB and Conference datasets were used to evaluate the effectiveness of the proposed feature selection system. We observed that when a subset of featur...

Co-clustering has been defined as a way to organize simultaneously subsets of instances and subsets of features in order to improve the clustering of both of them. In previous work [1], we proposed an efficient co-similarity measure allowing to simultaneously compute two similarity matrices between objects and features, each built on the basis of the other. Here we propose a generalization of this approach by introducing a notion of pseudo-norm and a pruning algorithm. Our experiments show that this new algorithm significantly improves the accuracy of the results when using either supervised or unsupervised feature selection data and that it outperforms other algorithms on various corpora.

Over the past few years, there has been a considerable spread of microarray technology in many biological patterns, particularly in those pertaining to cancer diseases like leukemia, prostate, colon cancer, etc. The primary bottleneck that one experiences in the proper understanding of such datasets lies in their dimensionality, and thus for an efficient and effective means of studying the same, a reduction in their dimension to a large extent is deemed necessary. This study is a bid to suggesting different algorithms and approaches for the reduction of dimensionality of such microarray datasets.This study exploits the matrix-like structure of such microarray data and uses a popular technique called Non-Negative Matrix Factorization (NMF) to reduce the dimensionality, primarily in the field of biological data. Classification accuracies are then compared for these algorithms.This technique gives an accuracy of 98%.

The application of soft computing in decision support system in disease prediction is one of the emerging interdisciplinary research areas in the field of computer science. Machine learning algorithms plays an important role in risk prediction of diseases. Attribute selection among the dataset is the key factor that influences prediction accuracy. Mathew's correlation coefficient performance metric is also taken into account. Particle swarm optimization algorithm is altered and applied for performing attribute selection. Improved fuzzy artificial neural network classifier performs the prediction task.

Feature selection is the task of choosing a small set out of a given set of features that capture the relevant properties of the data. In the context of supervised classification problems the relevance is determined by the given labels on the training data. A good choice of features is a key for building compact and accurate classifiers. In this paper we introduce a margin based feature selection criterion and apply it to measure the quality of sets of features. Using margins we devise novel selection algorithms for multi-class classification problems and provide theoretical generalization bound. We also study the well known Relief algorithm and show that it resembles a gradient ascent over our margin criterion. We apply our new algorithm to various datasets and show that our new Simba algorithm, which directly optimizes the margin, outperforms Relief.

Abstract In this paper, we introduce a new approach to learn dissimilarity for interactive search in content based image retrieval. In literature, dissimilarity is often learned via the feature space by feature selection, feature weighting or a parameterized function of the ...

Our previous works suggest that fractal texture feature is useful to detect pediatric brain tumor in multimodal MRI. In this study, we systematically investigate efficacy of using several different image features such as intensity, fractal texture, and level-set shape in segmentation of posterior-fossa (PF) tumor for pediatric patients. We explore effectiveness of using four different feature selection and three different segmentation techniques, respectively, to discriminate tumor regions from normal tissue in multimodal brain MRI. We further study the selective fusion of these features for improved PF tumor segmentation. Our result suggests that Kullback-Leibler divergence measure for feature ranking and selection and the expectation maximization algorithm for feature fusion and tumor segmentation offer the best results for the patient data in this study. We show that for T1 and fluid attenuation inversion recovery (FLAIR) MRI modalities, the best PF tumor segmentation is obtained using the texture feature such as multifractional Brownian motion (mBm) while that for T2 MRI is obtained by fusing level-set shape with intensity features. In multimodality fused MRI (T1, T2, and FLAIR), mBm feature offers the best PF tumor segmentation performance. We use different similarity metrics to evaluate quality and robustness of these selected features for PF tumor segmentation in MRI for ten pediatric patients.

Telemarketers of online job advertising firms face significant challenges understanding the advertising demands of small-sized enterprises. The effective use of data mining approach can offer e-recruitment companies an improved understanding of customers' patterns and greater insights of purchasing trends. However, prior studies on classifier built by data mining approach provided limited insights into the customer targeting problem of job advertising companies. In this paper we develop a single feature evaluator and propose an approach to select a desired feature subset by setting a threshold. The proposed feature evaluator demonstrates its stability and outstanding performance through empirical experiments in which real-world customer data of an e-recruitment firm are used. Practically, the findings together with the model may help telemarketers to better understand their customers. Theoretically, this paper extends existing research on feature selection for binary classification of imbalanced data.

An investigation into the feature extraction and selection of infant cry with asphyxia is presented in this paper. The feature of the cry signal was extracted using mel frequency cepstrum coefficient (MFCC) analysis and the significant coefficients were selected using orthogonal least square (OLS) algorithm. The effect of varying the number of MFCC filter banks on the feature selection was examined. It was found that the best set of coefficients could be achieved when 40 filter banks were used.

Women who have recovered from breast cancer (BC) always fear its recurrence. The fact that they have endured the painstaking treatment makes recurrence their greatest fear. However, with current advancements in technology, early recurrence prediction can help patients receive treatment earlier. The availability of extensive data and advanced methods make accurate and fast prediction possible. This research aims to compare the accuracy of a few existing data mining algorithms in predicting BC recurrence. It embeds a particle swarm optimization as feature selection into three renowned classifiers, namely, naive Bayes, K-nearest neighbor, and fast decision tree learner, with the objective of increasing the accuracy level of the prediction model.

Dimensionality refers to number of terms in a web page. While classifying web pages high dimensionality of web pages causes problem. The main objective of reducing dimensionality of web pages is improving the performance of classifier. Processing time and accuracy are two parameters which influence the performance of a classifier. To reduce the processing time, less informative and redundant terms have to be removed from web pages. This research describes hybrid approach for dimensionality reduction in web page classification using a rough set and naïve Bayesian method. Feature selection and dimensionality reduction methods are used for reducing the dimensionality. Information gain method is used as feature selection method. Rough set based Quick Reduct algorithm is used for dimensionality reduction. Naïve Bayesian method is used for classifying web pages to optimal predefined categories. Assignment of web pages to category is based on maximum posterior probability. Words remaining ...

We introduced a novel method employing a hierarchical domain ontology structure to extract features representing documents in our previous publication (Wang 2002). All raw words in the training documents are mapped to concepts in a concept hierarchy derived from the domain ontology. Based on these concepts, a concept hierarchy is established for the training document space, using is-a relationships defined in the domain ontology. An optimum concept set may be obtained by searching the concept hierarchy with an appropriate heuristic function. This may be used as the feature space to represent the training dataset. The proposed method aims to solve some drawbacks suffered by text classification algorithms and feature selection algorithms. In this paper, we conducted a series of experiments to compare our approach with other comparable feature-selection and feature-extraction methods. The results indicated that our approach has advantages in many aspects.

Pattern recognition generally requires that objects be described in terms of a set of measurable features. The selection and quality of the features representing each pattern have a considerable bearing on the success of subsequent pattern classification. Feature extraction is the process of deriving new features from the original features in order to reduce the cost of feature measurement, increase classifier efficiency, and allow higher classification accuracy. Many current feature extraction techniques involve linear transformations of the original pattern vectors to new vectors of lower dimensionality. While this is useful for data visualization and increasing classification efficiency, it does not necessarily reduce the number of features that must be measured since each new feature may be a linear combination of all of the features in the original pattern vector. Here, we present a new approach to feature extraction in which feature selection, feature extraction, and classifier training are performed simultaneously using a genetic algorithm. The genetic algorithm optimizes a vector of feature weights, which are used to scale the individual features in the original pattern vectors in either a linear or a nonlinear fashion. A masking vector is also employed to perform simultaneous selection of a subset of the features. We employ this technique in combination with the nearest neighbor classification rule, and compare the results with classical feature selection and extraction techniques, including sequential floating forward feature selection, and linear discriminant analysis. We also present results for the identification of favorable water-binding sites on protein surfaces, an important problem in biochemistry and drug design.

An intense research around classifier fusion in recent years revealed that combining performance strongly depends on careful selection of classifiers to be combined. Classifier performance depends, in turn, on careful selection of features, which could be further restricted by the subspaces of the data domain. On the other hand, there is already a number of classifier fusion techniques available and the choice of the most suitable method depends back on the selections made within classifier, features and data spaces. In all these multidimensional selection tasks genetic algorithms (GA) appear to be one of the most suitable techniques providing reasonable balance between searching complexity and the performance of the solutions found. In this work, an attempt is made to revise the capability of genetic algorithms to be applied to selection across many dimensions of the classifier fusion process including data, features, classifiers and even classifier combiners. In the first of the discussed models the potential for combined classification improvement by GA-selected weights for the soft combining of classifier outputs has been investigated. The second of the proposed models describes a more general system where the specifically designed GA is applied to selection carried out simultaneously along many dimensions of the classifier fusion process. Both, the weighted soft combiners and the prototype of the three-dimensional fusion-classifier-feature selection model have been developed and tested using typical benchmark datasets and some comparative experimental results are also presented. #

This paper studies the total variation regularization model with an L 1 fidelity term (TV-L 1 ) for decomposing an image into features of different scales. We first show that the images produced by this model can be formed from the minimizers of a sequence of decoupled geometry subproblems. Using this result we show that the TV-L 1 model is able to separate image features according to their scales, where the scale is analytically defined by the G-value. A number of other properties including the geometric and morphological invariance of the TV-L 1 model are also proved and their applications discussed.

We compared four automated methods for hippocampal segmentation using different machine learning algorithms (1) hierarchical AdaBoost, (2) Support Vector Machines (SVM) with manual feature selection, (3) hierarchical SVM with automated feature selection (Ada-SVM), and (4) a publicly available brain segmentation package (FreeSurfer). We trained our approaches using T1weighted brain MRI's from 30 subjects (10 normal elderly, 10 mild cognitive impairment (MCI), and 10 Alzheimer's disease (AD)), and tested on an independent set of 40 subjects (20 normal, 20 AD). Manually segmented gold standard hippocampal tracings were available for all subjects (training and testing). We assessed each approach's accuracy relative to manual segmentations, and its power to map AD effects. We then converted the segmentations into parametric surfaces to map disease effects on anatomy. After surface reconstruction, we computed significance maps, and overall corrected p-values, for the 3D profile of shape differences between AD and normal subjects. Our AdaBoost and Ada-SVM segmentations compared favorably with the manual segmentations and detected disease effects as well as FreeSurfer on the data tested. Cumulative pvalue plots, in conjunction with the False Discovery Rate method, were used to examine the power of each method to detect correlations with diagnosis and cognitive scores. We also evaluated how segmentation accuracy depended on the size of the training set, providing practical information for future users of this technique.

As the Internet grows at a phenomenal rate email systems has become a widely used electronic form of communication. Everyday, a large number of people exchange messages in this fast and inexpensive way. With the excitement on electronic commerce growing, the usage of email will increase more exponential. In this paper we present our research in developing general method for intrusion detection in email system Internet-based. The main ideas are to use data mining techniques to discover consistent and useful patterns of email system that can recognize anomalies and known intrusions.

Genetic algorithms have been created as an optimization strategy to be used especially when complex response surfaces do not allow the use of better-known methods (simplex, experimental design techniques, etc.). This paper shows that these algorithms, conveniently modified, can also be a valuable tool in solving the feature selection problem. The subsets of variables selected by genetic algorithms are generally more efficient than those obtained by classical methods of feature selection, since they can produce a better result by using a lower number of features.

Available in: http://www.redalyc.org/src/inicio/ArtPdfRed.jsp?iCve=43019328016 ... Redalyc Scientific Information System Network of Scientific Journals from Latin America, the Caribbean, Spain and ... A harmony search algorithm for clustering with feature selection ...

Extractive speech summarization, which purports to select an indicative set of sentences from a spoken document so as to succinctly represent the most important aspects of the document, has garnered much research over the years. In this paper, we cast extractive speech summarization as an ad-hoc information retrieval (IR) problem and investigate various language modeling (LM) methods for important sentence selection. The main contributions of this paper are four-fold. First, we explore a novel sentence modeling paradigm built on top of the notion of relevance, where the relationship between a candidate summary sentence and a spoken document to be summarized is discovered through different granularities of context for relevance modeling. Second, not only lexical but also topical cues inherent in the spoken document are exploited for sentence modeling. Third, we propose a novel clarity measure for use in important sentence selection, which can help quantify the thematic specificity of each individual sentence that is deemed to be a crucial indicator orthogonal to the relevance measure provided by the LM-based methods. Fourth, in an attempt to lessen summarization performance degradation caused by imperfect speech recognition, we investigate making use of different levels of index features for LM-based sentence modeling, including words, subword-level units, and their combination. Experiments on broadcast news summarization seem to demonstrate the performance merits of our methods when compared to several existing well-developed and/or state-of-the-art methods.

Robustness of feature selection techniques is a topic of recent interest, especially in high dimensional domains with small sample sizes, where selected feature subsets are subsequently analysed by domain experts to gain more insight into the problem modelled. In this work, we investigate the robustness of various feature selection techniques, and provide a general scheme to improve robustness using ensemble feature selection. We show that ensemble feature selection techniques show great promise for small sample domains, and provide more robust feature subsets than a single feature selection technique. In addition, we also investigate the effect of ensemble feature selection techniques on classification performance, giving rise to a new model selection strategy.

Many scientific problems can be formulated as classification tasks. Data that harbor relevant information are usually described by a large number of features. Frequently, many of these features are irrelevant for the class prediction. The efficient implementation of classification models requires identification of suitable combinations of features. The smaller number of fea- tures reduces the problem’s dimensionality and may result in higher classification perfor- mance. We developed DWFS, a web-based tool that allows for efficient selection of features for a variety of problems. DWFS follows the wrapper paradigm and applies a search strategy based on Genetic Algorithms (GAs). A parallel GA implementation examines and evaluates si- multaneously large number of candidate collections of features. DWFS also integrates various filtering methods that may be applied as a pre-processing step in the feature selection process. Furthermore, weights and parameters in the fitness function of GA can be adjusted according to the application requirements. Experiments using heterogeneous datasets from different bio- medical applications demonstrate that DWFS is fast and leads to a significant reduction of the number of features without sacrificing performance as compared to several widely used exist- ing methods. DWFS can be accessed online at www.cbrc.kaust.edu.sa/dwfs.

In this study, we propose a novel method for medical problem, it is the integration of particle swarm optimization (PSO) and decision tree (C4.5) named PSO + C4.5 algorithm. To evaluate the effectiveness of PSO + C4.5 algorithm, it is implemented on 5 different data sets of life sciences obtained from UCI machine learning databases. Moreover, the results of PSO + C4.5 implementation are compared to logistic regression (LR), back propagation neural network (BPNN), support vector machine (SVM), and decision tree (C4.5). The accuracy index shows that PSO + C4.5 algorithm outperforms the other methods.

Leucocytes in the blood of mammals form a powerful protective system against a wide range of dangerous pathogens. There are several types of immune cells that has specific role in the whole immune system. The number and type of immune cells alter in the disease state and identifying the type of immune cell provides information about a person's state of health. There are several immune cell subsets that are essentially morphologically identical and require external labeling to enable discrimination. Here we demonstrate the feasibility of using Wavelength Modulated Raman Spectroscopy (WMRS) with suitable machine learning algorithms as a label-free method to distinguish between different closely lying immune cell subset. Principal Component Analysis (PCA) was performed on WMRS data from single cells, obtained using confocal Raman microscopy for feature reduction, followed by Support Vector Machine (SVM) for binary discrimination of various cell subset, which yielded an accuracy >85%. The method was successful in discriminating between untouched and unfixed purified populations of CD4+CD3+ and CD8+CD3+ T lymphocyte subsets, and CD56+CD3-natural killer cells with a high degree of specificity. It was also proved sensitive enough to identify unique Raman signatures that allow clear discrimination between dendritic cell subsets, comprising CD303+CD45+ plasmacytoid and CD1c+CD141+ myeloid dendritic cells. The results of this study clearly show that WMRS is highly sensitive and can distinguish between cell types that are morphologically identical.

An automated approach to degradation analysis is proposed that uses a rotating machine's acoustic signal to determine Remaining Useful Life (RUL). High resolution spectral features are extracted from the acoustic data collected over the entire lifetime of the machine. A novel approach to the computation of Mutual Information based Feature Subset Selection is applied, to remove redundant and irrelevant features, that does not require class label boundaries of the dataset or spectral locations of developing defect to be known or pre-estimated. Using subsets of the feature space, multi-class linear and Radial Basis Function (RBF) Support Vector Machine (SVM) classifiers are developed and a comparison of their performance is provided. Performance of all classifiers is found to be very high, 85 to 98%, with RBF SVMs outperforming linear SVMs when a smaller number of features are used. As larger numbers of features are used for classification, the problem space becomes more linearly separable and the linear SVMs are shown to have comparable performance. A detailed analysis of the misclassifications is provided and an approach to better understand and interpret costly misclassifications is discussed. While defining class label boundaries using an automated k-means clustering algorithm improves performance with an accuracy of approximately 99%, further analysis shows that in 88% of all misclassifications the actual class of failure had the next highest probability of occurring. Thus, a system that incorporates probability distributions as a measure of confidence for the predicted RUL would provide additional valuable information for scheduling preventative maintenance.

Recently a large number of studies has been published in the area of Fractal Analysis. In this paper we review briefly the IFS (Iterated Function System) theory, and we show how this theoretical tool leads to new applications in Pattern Recognition and more precisely in the analysis of handwritten texts.

Identifying an appropriate set of predictors for the outcome of interest is a major challenge in clinical prediction research. The aim of this study is to show the application of some variable selection methods, usually used in data-mining, for an epidemiological study. We introduce here a systematic approach. The P-value-based method, usually used in epidemiological studies, and several filter and wrapper methods were implemented to select the predictors of diabetes among 55 variables in 803 pre-diabetic females, aged ≥20 years, followed for 10-12 years. To develop a logistic model, variables were selected from a train dataset and evaluated on the test dataset. The measures of AIC and AUC were used as performance criteria. We also implemented a full model with all 55 variables. We found that the worst and the best models were the full model and models based on the wrappers, respectively. Among filter methods, symmetrical uncertainty gave both the best AUC and AIC. Our experiment sh...

LASSO (Least Absolute Shrinkage and Selection Operator) is a useful tool to achieve the shrinkage and variable selection simultaneously. Since LASSO uses the L 1 penalty, the optimization should rely on the quadratic program (QP) or general non-linear program which is known to be computational intensive. In this paper, we propose a gradient descent algorithm for LASSO. Even though the final result is slightly less accurate, the proposed algorithm is computationally simpler than QP or non-linear program, and so can be applied to large size problems. We provide the convergence rate of the algorithm, and illustrate it with simulated models as well as real data sets.

We present MEDUSA, an integrative method for learning motif models of transcription factor binding sites by incorporating promoter sequence and gene expression data. We use a modern large-margin machine learning approach, based on boosting, to enable feature selection from the high-dimensional search space of candidate binding sequences while avoiding overfitting. At each iteration of the algorithm, MEDUSA builds a motif model whose presence in the promoter region of a gene, coupled with activity of a regulator in an experiment, is predictive of differential expression. In this way, we learn motifs that are functional and predictive of regulatory response rather than motifs that are simply overrepresented in promoter sequences. Moreover, MEDUSA produces a model of the transcriptional control logic that can predict the expression of any gene in the organism, given the sequence of the promoter region of the target gene and the expression state of a set of known or putative transcription factors and signaling molecules. Each motif model is either a k-length sequence, a dimer, or a PSSM that is built by agglomerative probabilistic clustering of sequences with similar boosting loss. By applying MEDUSA to a set of environmental stress response expression data in yeast, we learn motifs whose ability to predict differential expression of target genes outperforms motifs from the TRANSFAC dataset and from a previously published candidate set of PSSMs. We also show that MEDUSA retrieves many experimentally confirmed binding sites associated with environmental stress response from the literature.

Contemporary biological technologies produce extremely high-dimensional data sets from which to design classifiers, with 20,000 or more potential features being common place. In addition, sample sizes tend to be small. In such settings, feature selection is an inevitable part of classifier design. Heretofore, there have been a number of comparative studies for feature selection, but they have either considered settings with much smaller dimensionality than those occurring in current bioinformatics applications or constrained their study to a few real data sets. This study compares some basic feature-selection methods in settings involving thousands of features, using both model-based synthetic data and real data. It defines distribution models involving different numbers of markers (useful features) versus non-markers (useless features) and different kinds of relations among the features. Under this framework, it evaluates the performances of feature-selection algorithms for different distribution models and classifiers. Both classification error and the number of discovered markers are computed. Although the results clearly show that none of the considered feature-selection methods performs best across all scenarios, there are some general trends relative to sample size and relations among the features. For instance, the classifier-independent univariate filter methods have similar trends. Filter methods such as the t-test have better or similar performance with wrapper methods for harder problems. This improved performance is usually accompanied with significant peaking. Wrapper methods have better performance when the sample size is sufficiently large. ReliefF, the classifier-independent multivariate filter method, has worse performance than univariate filter methods in most cases; however, ReliefF-based wrapper methods show performance similar to their t-test-based counterparts.

This paper considers a class of hybrid (heterogeneous) ensembles purely composed of symbolic elements. In learning diagnostic rules from gene expressions they demonstrate a significant improvement of accuracy with a small number of ensemble elements. This makes them suitable for learning of understandable knowledge, leading to diagnosis and its explanation in original terms (attributes).

Recent work on extracting features of gaps in handwritten text allows a classification into inter-word and intraword classes using suitable classification techniques. In this paper, we apply 5 different supervised classification algorithms from the machine learning field on both the original dataset and a dataset with the best features selected using mutual information. The classifiers are compared by employing McNemar's test. We find that SVMs and MLPs outperform the other classifiers and that preprocessing to select features works well.

ABSTRACT Relief algorithms are successful attribute estimators. They are able to detect conditional dependencies between attributes and provide a unified view on the attribute estimation. In this paper, we propose a variant of ReliefF algorithm: ReliefMSS. We analyse the ReliefMSS parameters and compare ReliefF and ReliefMSS performances as regards the number of iterations, the number of random attributes, the noise effect, the number of nearest neighbours and the number of examples presented. We find that for the most of these parameters, ReliefMSS is better than ReliefF.

With the enabling Information Technology, disseminating every kind of information through the Internet became an enforcing condition for information providers. Now, it is more than being fashion and fun. Serving information on the Internet is a demand from the consumer as a commercial necessity. Internet allows all levels of society to access spatial data and even interactive mapping facilities that are available through the Web. Furthermore, Internet reshapes conventional characteristics of accessing, sharing, dissemination, visualization and analyzing conventions of spatial information handling. With today’s information technology support, user requirements determine remote geo-data processing as a need. In this paper a description is given of an application aiming at providing spatial data processing functions on the Internet is implemented using topographic data. In this framework, an architecture with the capability of services such as mapping, analyzing spatial data and dynami...

Nonnegative matrix factorization (NMF), a popular part-based representation technique, does not capture the intrinsic local geometric structure of the data space. Graph regularized NMF (GNMF) was recently proposed to avoid this limitation by regularizing NMF with a nearest neighbor graph constructed from the input data set. However, GNMF has two main bottlenecks. First, using the original feature space directly to construct the graph is not necessarily optimal because of the noisy and irrelevant features and nonlinear distributions of data samples. Second, one possible way to handle the nonlinear distribution of data samples is by kernel embedding. However, it is often difficult to choose the most suitable kernel. To solve these bottlenecks, we propose two novel graph-regularized NMF methods, AGNMF FS and AGNMF MK , by introducing feature selection and multiple-kernel learning to the graph regularized NMF, respectively. Instead of using a fixed graph as in GNMF, the two proposed methods learn the nearest neighbor graph that is adaptive to the selected features and learned multiple kernels, respectively. For each method, we propose a unified objective function to conduct feature selection/multi-kernel learning, NMF and adaptive graph regularization simultaneously. We further develop two iterative algorithms to solve the two optimization problems. Experimental results on two challenging pattern classification tasks demonstrate that the proposed methods significantly outperform state-of-the-art data representation methods.

This paper targets the automatic detection of personality traits in a meeting environment by means of audio and visual features; information about the relational context is captured by means of acoustic features designed to that purpose. Two personality traits are considered: Extraversion (from the Big Five) and the Locus of Control. The classification task is applied to thin slices of behaviour, in the form of 1-minute sequences. SVM were used to test the performances of several training and testing instance setups, including a restricted set of audio features obtained through feature selection. The outcomes improve considerably over existing results, provide evidence about the feasibility of the multimodal analysis of personality, the role of social context, and pave the way to further studies addressing different features setups and/or targeting different personality traits.