Binary Classification

One of the central problems in modern biology is to identify the complete set of interactions among the proteins in a cell. The structural interaction of proteins and their domains in networks is one of the most basic molecular mechanisms for biological cells. Structural evidence indicates that, interacting pairs of close homologs usually interact in the same way. In this article, we make use of both homology and inter-domain linker region knowledge to predict interaction between protein pairs solely by amino acid sequence information. High quality core set of 150 yeast proteins obtained from the Database of Interacting Proteins (DIP) was considered to test the accuracy of the proposed method. The strongest prediction of the method reached over 70% accuracy. These results show great potential for the proposed method.

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.

Retinal blood vessel assessment plays an important role in the diagnosis of ophthalmic pathologies. The use of digital images for this purpose enables the application of a computerized approach and has fostered the development of multiple methods for automated vascular tree segmentation. Metrics based on contingency tables for binary classification have been widely used for evaluating the performance of these algorithms. Metrics from this family are based on the measurement of a success or failure rate in the detected pixels, obtained by means of pixel-to-pixel comparison between the automated segmentation and a manually-labeled reference image. Therefore, vessel pixels are not considered as a part of a vascular structure with specific features. This paper contributes a function for the evaluation of global quality in retinal vessel segmentations. This function is based on the characterization of vascular structures as connected segments with measurable area and length. Thus, its design is meant to be sensitive to anatomical vascularity features. Comparison of results between the proposed function and other general quality evaluation functions shows that this proposal renders a high matching degree with human quality perception. Therefore, it can be used to enhance quality evaluation in retinal vessel segmentations, supplementing the existing functions. On the other hand, from a general point of view, the applied concept of measuring descriptive properties may be used to design specialized functions aimed at segmentation quality evaluation in other complex structures.

Cohen's kappa index is reformulated for multiple classifications based on exchangeable random variables. It is found that kappa is between 0 and I inclusive, Two characterizations for kappa are stated in terms of the relationship between such random variables. Within the normal test score model, kappa increases with test reliability and test length. Furthermore, when based on binary' classifications, kappa is an inverse U-shaped function of the cutoff score. These trends also bold for the beta-binomial test score model.

The authors describe a model-based kappa statistic for binary classifications which is interpretable in the same manner as Scott's pi and Cohen's kappa, yet does not suffer from the same flaws. They compare this statistic with the data-driven and population-based forms of Scott's pi in a population-based setting where many raters and subjects are involved, and inference regarding the underlying diagnostic procedure is of interest. The authors show that Cohen's kappa and Scott's pi seriously underestimate agreement between experts classifying subjects for a rare disease; in contrast, the new statistic is robust to changes in prevalence. The performance of the three statistics is illustrated with simulations and prostate cancer data.

This paper investigates the influence of different page features on the ranking of search engine results. We use Google (via its API) as our testbed and analyze the result rankings for several queries of different categories using statistical methods. We reformulate the problem of learning the underlying, hidden scores as a binary classification problem. To this problem we then apply both linear and non-linear methods. In all cases, we split the data into a training set and a test set to obtain a meaningful, unbiased estimator for the quality of our predictor. Although our results clearly show that the scoring function cannot be approximated well using only the observed features, we do obtain many interesting insights along the way and discuss ways of obtaining a better estimate and main limitations in trying to do so.

Due to the enormous amount of data and opinions being produced, shared and transferred everyday across the internet and other media, Sentiment analysis has become vital for developing opinion mining systems. This paper introduces a developed classification sentiment analysis using deep learning networks and introduces comparative results of different deep learning networks. Multilayer Perceptron (MLP) was developed as a baseline for other networks results. Long short-term memory (LSTM) recurrent neural network, Convolutional Neural Network (CNN) in addition to a hybrid model of LSTM and CNN were developed and applied on IMDB dataset consists of 50K movies reviews files. Dataset was divided to 50% positive reviews and 50% negative reviews. The data was initially pre-processed using Word2Vec and word embedding was applied accordingly. The results have shown that, the hybrid CNN_LSTM model have outperformed the MLP and singular CNN and LSTM networks. CNN_LSTM have reported the accuracy of 89.2% while CNN has given accuracy of 87.7%, while MLP and LSTM have reported accuracy of 86.74% and 86.64 respectively. Moreover, the results have elaborated that the proposed deep learning models have also outperformed SVM, Naïve Bayes and RNTN that were published in other works using English datasets.

The purpose of this paper is to propose a new binary classification method for predicting corporate failure based on genetic algorithm, and to validate its prediction power through empirical analysis. Establishing virtual companies representing bankrupt companies and non-bankrupt ones, respectively, the proposed method measures the similarity between the virtual companies and the subject for prediction, and classifies the subject into either bankrupt or non-bankrupt one. The values of the classification variables of the virtual companies and the weights of the variables are determined by the proper model to maximize the hit ratio of training data set using genetic algorithm. In order to test the validity of the proposed method, we compare its prediction accuracy with those of other existing methods such as multi-discriminant analysis, logistic regression, decision tree, and artificial neural network, and it is shown that the binary classification method we propose in this paper can serve as a promising alternative to the existing methods for bankruptcy prediction.

Several real problems involve the classification of data into categories or classes. Given a data set containing data whose classes are known, Machine Learning algorithms can be employed for the induction of a classifier able to predict the class of new data from the same domain, performing the desired discrimination. Some learning techniques are originally conceived for the solution of problems with only two classes, also named binary classification problems. However, many problems require the discrimination of examples into more than two categories or classes. This paper presents a survey on the main strategies for the generalization of binary classifiers to problems with more than two classes, known as multiclass classification problems. The focus is on strategies that decompose the original multiclass problem into multiple binary subtasks, whose outputs are combined to obtain the final prediction.

Healthcare expenditure is a growing concern in the US. In 2012, the total US annual healthcare expenditure reached $2.8 trillion. As a percentage of GDP, it is the highest among all the nations worldwide. In this study we investigate the variability of patient healthcare expenses year-onyear, depending on the different medical conditions patients get diagnosed with, the prescription drugs they consume and the demographic variables. We work with anonymized but publicly available Medicare data, which has more than 114,000 beneficiaries and more than 12,400 features. We address the problem of accurately predicting which beneficiaries' inpatient claim amounts increased between 2008 and 2009, using an ensemble of six different classification algorithms. We achieved a sensitivity (recall) of 80%, an overall accuracy of 77.56% and a precision of 76.46% on the test data set. We demonstrate its benefits to the healthcare stakeholders: the insurance providers for projecting cost and revenue more accurately, the high-risk patients for choosing the right insurance plan, and the healthcare providers for deciding which patients need additional monitoring. Our research shows that kidney conditions, COPD (chronic obstructive pulmonary disease), hypertension, stroke/transient ischemic attack, cancer and osteoporosis are among the most influential conditions behind expenditure increase.

Convolutional Neural Networks (CNNs) are a subset of Supervised Learning class of algorithms that are very similar to regular Neural Networks and aim to find an optimal predictive model that assigns the input variable to the correct label.

In contrast to the Multilayer Perceptron Architecture (MLP) that uses fully connected network layers, a CNN does not need to provide information of the entire feature space to all hidden layer nodes, but instead it breaks the input matrix into regions and then connects each region to a single hidden node. With this regional breakdown and assignment of small local groups of features to different hidden nodes, CNNs are performing very well for image recognition tasks.

On the other hand, a Support Vector Machine classifier tries to separate the data into K classes by maximizing the distance between the differently labeled data. If the data are not linearly separable, then by using an appropriate kernel function we can map the data into a higher dimension where they happen to be linearly separable and we find the linear boundary there. Finally, we transform that linear boundary back to the original lower dimensions and we get a non-linear separator. In this project we are going to replace the standard sigmoid activation function of the penultimate layer of the network with a linear Support Vector Machine classifier and investigate performance differences. We are going to implement the standard CNN architecture as benchmark model and see how it compares with a Deep Learning SVC so that we choose the best model to implement the final solution.

The problem of testing equivalence of two ROC curves is addressed. A transformation of corresponding ROC curves, which motivates a test statistic based on a distance of two empirical quantile processes, is suggested, its asymptotic distribution found and a simulation scheme proposed that enables us to find critical values.

Due to the enormous amount of data and opinions being produced, shared and transferred everyday across the internet and other media, Sentiment analysis has become vital for developing opinion mining systems. This paper introduces a developed classification sentiment analysis using deep learning networks and introduces comparative results of different deep learning networks. Multilayer Perceptron (MLP) was developed as a baseline for other networks results. Long short-term memory (LSTM) recurrent neural network, Convolutional Neural Network (CNN) in addition to a hybrid model of LSTM and CNN were developed and applied on IMDB dataset consists of 50K movies reviews files. Dataset was divided to 50% positive reviews and 50% negative reviews. The data was initially pre-processed using Word2Vec and word embedding was applied accordingly. The results have shown that, the hybrid CNN_LSTM model have outperformed the MLP and singular CNN and LSTM networks. CNN_LSTM have reported the accuracy of 89.2% while CNN has given accuracy of 87.7%, while MLP and LSTM have reported accuracy of 86.74% and 86.64 respectively. Moreover, the results have elaborated that the proposed deep learning models have also outperformed SVM, Naïve Bayes and RNTN that were published in other works using English datasets.

The primary purposes of binary classification is performance optimization since even the slightest prediction improvements can have signification implications for each field application. Finding the most effective separation of classes is the key success indicator in how well each model performs. When modeling prediction fails, the data entry is placed in either a false positive (incorrectly predicted) or negative (incorrectly failed to predict) category. The consequences for failing to correctly categorize can range from wasting resources and missed opportunities to serious harm caused in medical diagnoses. Optimizing classification performance focuses on reducing misclassification.

Improvements can be identified after applying metrics then in finding ways to increase accuracy and reduce misclassification. Accuracy of classification and AUCROC, area under the Receiver Operating curve, are the most commonly applied metrics, however identifying insights into misclassification requires reviewing Type I and II error metrics. Ensemble methods (bagging, boosting and stacking) are a creative means of resampling and will be utilized to improve the performance of base learners in stacked models. The German and Australian credit risk scoring datasets were run through 9 diverse algorithms, as well as adding bagging and boosting in ensemble stacked models based on strong and weak learner correlations. The results were ranked based on the overall optimal performance.

The Support Vector Machine (SVM) algorithm ranked best for both datasets with no false positives and negatives. The stacked models were in 16 of the top 17 ranked models,however the ranking of specific models was inconsistent. This study found that the stacking ensemble premise did not have a clear delineation in performance improvement in combining the weakest and strongest learners together, however stacking itself proved very successful. The premise is that combining the weakest learners, least correlated, should be more effective than the strong learners since they have the most to benefit from reweighting for resampling that promotes the weakest and demotes the stronger learners helping to focus on the weakest algorithms fixing their misclassifications.

The conclusion regarding model choice is that no assumptions could be made regarding algorithm performance and that empirical testing is needed for model selection.

The key implication of optimizing classification performance is the reduction of misclassification errors and thereby enabling greater classification accuracy which translates into gaining better insights from the risk score modeling process that can be effectively prescribed back in an organization. One of the most important areas of future work continues to be finding solutions to class balancing which improve the quality of performance metrics that are inherently skewed.

Keywords: business risk assessment, classification performance optimization, credit risk scoring, risk scoring analytics

Risk Scoring Classification Performance Optimization. Available from: https://www.researchgate.net/publication/320011901_Risk_Scoring_Classification_Performance_Optimization [accessed Sep 24, 2017].

This tutorial provides a concise overview of support vector machines and different closely related techniques for pattern classification. The tutorial starts with the formulation of support vector machines for classification. The method of least squares support vector machines is explained. Approaches to retrieve a probabilistic interpretation are covered and it is explained how the binary classification techniques can be extended to multi-class methods. Kernel logistic regression, which is closely related to iteratively weighted least squares support vector machines, is discussed. Different practical aspects of these methods are addressed: the issue of feature selection, parameter tuning, unbalanced data sets, model evaluation and statistical comparison. The different concepts are illustrated on three real-life applications in the field of metabolomics, genetics and proteomics.

In an era of strong customer relationship management (CRM) emphasis, firms strive to build valuable relationships with their existing customer base. In this study, we attempt to better understand three important measures of customer outcome: next buy, partial-defection and customers' profitability evolution. By means of random forests techniques we investigate a broad set of explanatory variables, including past customer behavior, observed customer heterogeneity and some typical variables related to intermediaries. We analyze a real-life sample of 100,000 customers taken from the data warehouse of a large European financial services company. Two types of random forests techniques are employed to analyze the data: random forests are used for binary classification, whereas regression forests are applied for the models with linear dependent variables. Our research findings demonstrate that both random forests techniques provide better fit for the estimation and validation sample compared to ordinary linear regression and logistic regression models. Furthermore, we find evidence that the same set of variables have a different impact on buying versus defection versus profitability behavior. Our findings suggest that past customer behavior is more important to generate repeat purchasing and favorable profitability evolutions, while the intermediary's role has a greater impact on the customers' defection proneness. Finally, our results demonstrate the benefits of analyzing different customer outcome variables simultaneously, since an extended investigation of the next buy-partial-defection-customer profitability triad indicates that one cannot fully understand a particular outcome without understanding the other related behavioral outcome variables.

Background Apraxia in patients with stroke may be overlooked, as clumsiness and deficient gestural communication are often attributed to frequently coexisting sensorimotor deficits and aphasia. Early and reliable detection of apraxia by a bedside test is relevant for functional outcome in patients with stroke. The present study was aimed at constructing a new bedside screening test for apraxia, called the Apraxia Screen of TULIA (AST), based on the comprehensive standardised Test for Upper-Limb Apraxia (TULIA). Methods First, an item-reduction analysis of the TULIA (48 gestures) was performed, based on the methods of classical test theory and on a larger sample of patients with stroke (n¼133) and matched healthy controls (n¼50). Stepwise elimination of items resulted in a set of 12 items, demonstrating high internal consistency (Cronbach alpha¼0.92). The six-point scoring method of the TULIA was dichotomised to the score levels pass and fail. In the second part of this study the validity of the AST was assessed prospectively in a new cohort of patients with stroke (n¼31) by using the Pearson correlation analysis and binary classification display with the TULIA. Results and discussion Validation of the 12-item AST with the TULIA showed a remarkable diagnostic reliability with high specificity, sensitivity and positive predictive value, for the presence and severity of apraxia. The AST is shown to be a reliable and valid bedside test in patients with stroke, allowing a straightforward assessment of apraxia within a few minutes.

This paper addresses the prediction of epileptic seizures from the online analysis of EEG data. This problem is of paramount importance for the realization of monitoring/control units to be implanted on drug-resistant epileptic patients. The proposed solution relies in a novel way on autoregressive modeling of the EEG time series and combines a least-squares parameter estimator for EEG feature extraction along with a support vector machine (SVM) for binary classification between preictal/ictal and interictal states. This choice is characterized by low computational requirements compatible with a real-time implementation of the overall system. Moreover, experimental results on the Freiburg dataset exhibited correct prediction of all seizures (100% sensitivity) and, due to a novel regularization of the SVM classifier based on the Kalman filter, also a low false alarm rate.

—A Worm is a standalone malware that replicates itself and does not need a host to propagate. The prevailing detection approach uses fixed size sequence of characters extracted from the worm as a signature. Although very fast, the signature detection approach is ineffective for zero day attacks. Building upon our previous work to use data mining techniques for malware detection, we present a comparison of variable length instruction sequence to fixed size n-grams to detect worms. We define a sequence as a series of instructions till a branch is ennountered. Our results indicated that using the variable length sequences as features to a machine learning classifier provided better results than using fixed length n-grams. We built and compared several tree based classifiers and were able to achieve 95.6% detection rate on novel worms using the variable length instruction sequences.

Bankruptcy prediction has been a topic of research for decades, both within the financial and the academic world. The implementations of international financial and accounting standards, such as Basel II and IFRS, as well as the recent credit crisis, have accentuated this topic even further. This paper describes both regularized and non-linear kernel variants of traditional discriminant analysis techniques, such as logistic regression, Fisher discriminant analysis (FDA) and quadratic discriminant analysis (QDA). Next to a systematic description of these variants, we contribute to the literature by introducing kernel QDA and providing a comprehensive benchmarking study of these classification techniques and their regularized and kernel versions for bankruptcy prediction using 10 real-life data sets. Performance is compared in terms of binary classification accuracy, relevant for evaluating yes/no credit decisions and in terms of classification accuracy, relevant for pricing differentiated credit granting. The results clearly indicate the significant improvement for kernel variants in both percentage correctly classified (PCC) test instances and area under the ROC curve (AUC), and indicate that bankruptcy problems are weakly non-linear. On average, the best performance is achieved by LSSVM, closely followed by kernel quadratic discriminant analysis. Given the high impact of small improvements in performance, we show the relevance and importance of considering kernel techniques within this setting. Further experiments with backwards input selection improve our results even further. Finally, we experimentally investigate the relative ranking of the different categories of variables: liquidity, solvency, profitability and various, and as such provide new insights into the relative importance of these categories for predicting financial distress.

Two enhancements are proposed to the application and theory of support vector machines. The first is a method of multicategory classification based on the binary classification version of the support vector machine (SVM). The method, which is called the -ary SVM, represents each category in binary format, and to each bit of that representation is assigned a conventional SVM. This approach requires only log 2 ( ) SVMs, where is the number of classes. We give an example of classification on an octaphase-shift-keying (8-PSK) pattern space to illustrate main concepts.

Support vector machine (SVM) is a well sound learning method and a robust classification procedure. Choosing a suitable kernel function in SVM is crucial for obtaining good performance; the difficulty is how to choose a suitable data transformation for the given problem. To this end, multiple kernel matrices, each of them corresponding to a given similarity measure, can be linearly combined. In this paper, the optimal kernel matrix, obtained as linear combination of known kernel matrices, is generated using a semidefinite programming approach. A suitable model formulation assures that the obtained kernel matrix is positive semidefinite and is optimal with respect to the dataset under consideration. The proposed approach has been applied to some very important medical diagnostic decision making problems and the results obtained by carrying out preliminary numerical experiments demonstrated the effectiveness of the proposed solution approach.

a b s t r a c t Support vector machines (SVMs) were originally formulated for the solution of binary classification problems. In multiclass problems, a decomposition approach is often employed, in which the multiclass problem is divided into multiple binary subproblems, whose results are combined. Generally, the performance of SVM classifiers is affected by the selection of values for their parameters. This paper investigates the use of genetic algorithms (GAs) to tune the parameters of the binary SVMs in common multiclass decompositions. The developed GA may search for a set of parameter values common to all binary classifiers or for differentiated values for each binary classifier.

Learning algorithms can suffer a performance bias when data sets only have a small number of training examples for one or more classes. In this scenario learning methods can produce the deceptive appearance of "good looking" results even when classification performance on the important minority class can be poor. This paper compares two Genetic Programming (GP) approaches for classification with unbalanced data. The first focuses on adapting the fitness function to evolve classifiers with good classification ability across both minority and majority classes. The second uses a multi-objective approach to simultaneously evolve a Pareto front (or set) of classifiers along the minority and majority class tradeoff surface. Our results show that solutions with good classification ability were evolved across a range of binary classification tasks with unbalanced data.

Words and phrases acquire meaning from the way they are used in society, from their relative semantics to other words and phrases. For computers the equivalent of 'society' is 'database,' and the equivalent of 'use' is 'way to search the database.' We present a new theory of similarity between words and phrases based on information distance and Kolmogorov complexity. To fix thoughts we use the world-wide-web as database, and Google as search engine. The method is also applicable to other search engines and databases. This theory is then applied to construct a method to automatically extract similarity, the Google similarity distance, of words and phrases from the world-wideweb using Google page counts. The world-wide-web is the largest database on earth, and the context information entered by millions of independent users averages out to provide automatic semantics of useful quality. We give applications in hierarchical clustering, classification, and language translation. We give examples to distinguish between colors and numbers, cluster names of paintings by 17th century Dutch masters and names of books by English novelists, the ability to understand emergencies, and primes, and we demonstrate the ability to do a simple automatic English-Spanish translation. Finally, we use the WordNet database as an objective baseline against which to judge the performance of our method. We conduct a massive randomized trial in binary classification using support vector machines to learn categories based on our Google distance, resulting in an a mean agreement of 87% with the expert crafted WordNet categories.

This paper investigates the feature subset selection problem for the binary classification problem using logistic regression model. We developed a modified discrete particle swarm optimization (PSO) algorithm for the feature subset selection problem. This approach embodies an adaptive feature selection procedure which dynamically accounts for the relevance and dependence of the features included the feature subset. We compare the proposed methodology with the tabu search and scatter search algorithms using publicly available datasets. The results show that the proposed discrete PSO algorithm is competitive in terms of both classification accuracy and computational performance.

In this paper we present a novel approach and a new machine learning problem, called Supervised Novelty Detection (SND). This problem extends the One-Class Support Vector Machine setting for binary classification while keeping the nice properties of novelty detection problem at hand. To tackle this we approach binary classification from a new perspective using two different estimators and a coupled regularization term. It involves optimization over a different objective and a doubled set of Lagrange multipliers. One might consider our approach as a joint estimation of the support for different probability distributions per class where an ultimate goal is to separate classes with the largest possible angle between the normal vectors to the decision hyperplanes in the feature space. Regarding an obvious novelty of our problem we report and compare the results along the lines of standard C-SVM, LS-SVM and One-Class SVM. Experiments have demonstrated promising results that validate the usefulness of the proposed method.

This paper presents an approach to the joint optimization of neural network structure and weights which can take advantage of backpropagation as a specialized decoder. The approach is applied to binary classification of brain waves in the context of brain-computer interfaces.

We show that the Confusion Entropy, a measure of performance in multiclass problems has a strong (monotone) relation with the multiclass generalization of a classical metric, the Matthews Correlation Coefficient. Analytical results are provided for the limit cases of general no-information (n-face dice rolling) of the binary classification. Computational evidence supports the claim in the general case.

Signalized intersections are accident-prone areas especially for rear-end crashes due to the fact that the diversity of the braking behaviors of drivers increases during the signal change. The objective of this article is to improve knowledge of the relationship between rear-end crashes occurring at signalized intersections and a series of potential traffic risk factors classified by driver characteristics, environments, and vehicle types.

We participated (as Team 9) in the Article Classification Task of the Biocreative II.5 Challenge: binary classification of fulltext documents relevant for protein-protein interaction. We used two distinct classifiers for the online and offline challenges: (1) the lightweight Variable Trigonometric Threshold (VTT) linear classifier we successfully introduced in BioCreative 2 for binary classification of abstracts, and (2) a novel Naive Bayes classifier using features from the citation network of the relevant literature. We supplemented the supplied training data with full-text documents from the MIPS database. The lightweight VTT classifier was very competitive in this new full-text scenario: it was a top performing submission in this task, taking into account the rank product of the Area Under the interpolated precision and recall Curve, Accuracy, Balanced F-Score, and Matthew's Correlation Coefficient performance measures. The novel citation network classifier for the biomedical text mining domain, while not a top performing classifier in the challenge, performed above the central tendency of all submissions and therefore indicates a promising new avenue to investigate further in bibliome informatics.

Misbehavior detection in vehicular ad hoc networks (VANETs) is performed to improve the traffic safety and driving accuracy. All the nodes in the VANETs communicate to each other through message logs. Malicious nodes in the VANETs can cause inevitable situation by sending message logs with tampered values. In this work, various machine learning algorithms are used to detect the primarily five types of attacks namely, constant attack, constant offset attack, random attack, random offset attack, and eventual attack. Firstly, each attack is detected by different machine learning algorithms using binary classification. Then, the new procedure is created to do the multi classification of the attacks on best chosen algorithm from different machine learning techniques. The highest accuracy in case of binary classification is obtained with Naïve Bayes (100%), decision tree (100%), and random forest (100%) in type1 attack, decision tree (100%) in type2 attack, and random forest (98.03%, 95.56%, and 95.55%) in Type4, Type8 and Type16 attack respectively. In case of new procedure for multiclassification, the highest accuracy is obtained with random forest (97.62%) technique. For this work, VeReMi dataset (a public repository for the malicious node detection in VANETs) is used.

We define voice activity detection (VAD) as a binary classification problem and solve it using the support vector machine (SVM). Challenges in SVM-based approach include selection of representative training segments, selection of features, normalization of the features, and post-processing of the frame-level decisions. We propose to construct a SVM-VAD using MFCC features because they capture the most relevant information of speech, and they are widely used in speech and speaker recognition making the proposed method easy to integrate with existing applications. Practical usability is our driving motivation: the proposed SVM-VAD should be easily adapted into new conditions. Index Terms: voice activity detection (VAD), machine learning, support vector machine (SVM)

We explored the use of a fiber-optic probe for in vivo fluorescence spectroscopy of breast tissues during percutaneous image-guided breast biopsy. A total of 121 biopsy samples with accompanying histological diagnosis were obtained clinically and investigated in this study. The tissue spectra were analyzed using partial least-squares analysis and represented using a set of principal components (PCs) with dramatically reduced data dimension. For nonmalignant tissue samples, a set of PCs that account for the largest amount of variance in the spectra displayed correlation with the percent tissue composition. For all tissue samples, a set of PCs was identified using a Wilcoxon rank-sum test as showing statistically significant differences between: 1) malignant and fibrous/benign; 2) malignant and adipose; and 3) malignant and nonmalignant breast samples. These PCs were used to distinguish malignant from other nonmalignant tissue types using a binary classification scheme based on both linear and nonlinear support vector machine (SVM) and logistic regression (LR). For the sample set investigated in this study, the SVM classifier provided a crossvalidated sensitivity and specificity of up to 81% and 87%, respectively, for discrimination between malignant and fibrous/benign samples, and up to 81% and 81%, respectively, for discriminating between malignant and adipose samples. Classification based on LR was used to generate receiver operator curves with an area under the curve (AUC) of 0.87 for discriminating malignant versus fibrous/benign tissues, and an AUC of 0.84 for discriminating malignant from adipose tissue samples. This study demonstrates the feasibility of performing fluorescence spectroscopy during clinical core needle breast biopsy, and the potential of this technique for identifying breast malignancy in vivo.

We have proposed a hybrid SVM based decision tree to speedup SVMs in its testing phase for binary classification tasks. While most existing methods addressed towards this task aim at reducing the number of support vectors, we have focused on reducing the number of test datapoints that need SVM's help in getting classified. The central idea is to approximate the decision boundary of SVM using decision trees. The resulting tree is a hybrid tree in the sense that it has both univariate and multivariate (SVM) nodes. The hybrid tree takes SVM's help only in classifying crucial datapoints lying near decision boundary; remaining less crucial datapoints are classified by fast univariate nodes. The classification accuracy of the hybrid tree is guaranteed by tuning a threshold parameter. Extensive computational comparisons on 19 publicly available datasets indicate that the proposed method achieves significant speedup when compared to SVMs, without any compromise in classification accuracy.

The properties of acoustic speech have previously been investigated as possible cues for depression in adults. However, these studies were restricted to small populations of patients and the speech recordings were made during patients' clinical interviews or fixed-text reading sessions. Symptoms of depression often first appear during adolescence at a time when the voice is changing, in both males and females, suggesting that specific studies of these phenomena in adolescent populations are warranted. This study investigated acoustic correlates of depression in a large sample of 139 adolescents (68 clinically depressed and 71 controls). Speech recordings were made during naturalistic interactions between adolescents and their parents. Prosodic, cepstral, spectral, and glottal features, as well as features derived from the Teager energy operator (TEO), were tested within a binary classification framework. Strong gender differences in classification accuracy were observed. The TEO-based features clearly outperformed all other features and feature combinations, providing classification accuracy ranging between 81%-87% for males and 72%-79% for females. Close, but slightly less accurate, results were obtained by combining glottal features with prosodic and spectral features (67%-69% for males and 70%-75% for females). These findings indicate the importance of nonlinear mechanisms associated with the glottal flow formation as cues for clinical depression.

Ensembles that combine the decisions of classifiers generated by using perturbed versions of the training set where the classes of the training examples are randomly switched can produce a significant error reduction, provided that large numbers of units and high class switching rates are used. The classifiers generated by this procedure have statistically uncorrelated errors in the training set. Hence, the ensembles they form exhibit a similar dependence of the training error on ensemble size, independently of the classification problem. In particular, for binary classification problems, the classification performance of the ensemble on the training data can be analysed in terms of a Bernoulli process. Experiments on several UCI datasets demonstrate the improvements in classification accuracy that can be obtained using these class-switching ensembles.

In many real-world binary classification tasks (e.g. detection of certain objects from images), an available dataset is imbalanced, i.e., it has much less representatives of a one class (a minor class), than of another. Generally, accurate prediction of the minor class is crucial but it's hard to achieve since there is not much information about the minor class. One approach to deal with this problem is to preliminarily resample the dataset, i.e., add new elements to the dataset or remove existing ones. Resampling can be done in various ways which raises the problem of choosing the most appropriate one. In this paper we experimentally investigate impact of resampling on classification accuracy, compare resampling methods and highlight key points and difficulties of resampling.

INTRODUCTION: One limitation for the study of Uromyces appendiculatus (Pers.: Pers.) Unger (sin. U. phaseoli (Reben) Wint.), the causal agent of rust in common bean (Phaseolus vulgaris L.), was the system used for the classification of physiological races. This system considered 19 differential cultivars and a symptom evaluation scale with 37 reaction degrees (Stavely et al., 1983; Stavely, 1984). To facilitate this process, in the "3rd Bean Rust International Workshop", held in South Africa, in 2002, a new differential series was defined containing six Andean and six Mesoamerican cultivars (Table 1). The symptom evaluation scale was organized in only two classes of reactions, resistant and susceptible, and also a binary system was proposed to designate each race (Steadman et al., 2002). The proposed system standardizes the classification of U. appendiculatus physiologic races and facilitates the exchange of information and the cooperative use of the results obtained by di...

Forecasting applications on the stock market attract much interest from researchers in the artificial intelligence field. The problem tackled in this study concerns predicting the direction of change of stock price indices, formulated in terms of binary classification. We use gene expression programming to evolve pools of binary classifiers and investigate several approaches to construct ensembles based on them. We compare the performance of the obtained classifiers with those of Naive Bayes, Support Vector Machines, Multilayer Perceptron, Decision Random Forrest. The experiments performed on real-world stock market data show that the ensembles of GEPevolved classifier models are competitive to classifiers trained by state-of-the-art machine learning methods.

In several computer-aided diagnosis (CAD) applications of image processing, there is no sufficiently sensitive and specific method for determining what constitutes a normal versus an abnormal classification of a chest radiograph. In the case of lung nodule detection or in classifying the perfusion of pneumoconiosis, multiple radiograph readers (radiologists) are asked to examine and score specific regions of interest (ROIs). The readers provide size, shape and perfusion grades for the presence of opacities in each region and then use all the ROI grades to classify the lung as normal or abnormal. The combined grades from all readers are then used to arrive at a consensus normal or abnormal classification. In this paper, using area under the ROC curve, we evaluate new mathematical models that are based on mathematical statistics, logic functions, and several statistical classifiers to analyze reader performance in grading chest radiographs for pneumoconiosis as the first step toward applying this technique to early detection of nodules found in lung cancer. In pneumoconiosis, rounded opacities are on the order of 1-10 mm in size, while lung nodules are often not diagnosed until they reach a size on the order of 1 cm.

The paper proposes a different approach to data modeling. Analogous to the rejection method, where the misclassifications are removed and manually evaluated, we focus here on difficult to distinguish cases for binary classification. Such cases are further explored and information granulation is conducted based on the idea to stretch out the interesting intervals. This refinement model adopts the concept from database theory where the n x n relations are resolved through additional internal attribute connections. We introduce an integration of target functions for such cases. The achieved experimental results from a test dataset are described and future work for user assistance in knowledge modelling is presented.

We consider two on-line learning frameworks: binary classification through linear threshold functions and linear regression. We study a family of on-line algorithms, called p-norm algorithms, introduced by Grove, Littlestone and Schuurmans in the context of deterministic binary classification. We show how to adapt these algorithms for use in the regression setting, and prove worst-case bounds on the square loss, using a technique from Kivinen and Warmuth. As pointed out by Grove, et al., these algorithms can be made to approach a version of the classification algorithm Winnow as p goes to infinity; similarly they can be made to approach the corresponding regression algorithm EG in the limit. Winnow and EG are notable for having loss bounds that grow only logarithmically in the dimension of the instance space. Here we describe another way to use the p-norm algorithms to achieve this logarithmic behavior. With the way to use them that we propose, it is less critical than with Winnow and EG to retune the parameters of the algorithm as the learning task changes. Since the correct setting of the parameters depends on characteristics of the learning task that are not typically known a priori by the learner, this gives the p-norm algorithms a desireable robustness. Our elaborations yield various new loss bounds in these on-line settings. Some of these bounds improve or generalize known results. Others are incomparable.

Machine learning algorithms such as genetic programming (GP) can evolve biased classifiers when data sets are unbalanced. Data sets are unbalanced when at least one class is represented by only a small number of training examples (called the minority class) while other classes make up the majority. In this scenario, classifiers can have good accuracy on the majority class but very poor accuracy on the minority class(es) due to the influence that the larger majority class has on traditional training criteria in the fitness function. This paper aims to both highlight the limitations of the current GP approaches in this area and develop several new fitness functions for binary classification with unbalanced data. Using a range of real-world classification problems with class imbalance, we empirically show that these new fitness functions evolve classifiers with good performance on both the minority and majority classes. Our approaches use the original unbalanced training data in the GP learning process, without the need to artificially balance the training examples from the two classes (e.g., via sampling).