Search

Abstract Multilabel classification is an emergent data mining task with a broad range of real world applications. Learning from imbalanced multilabel data is being deeply studied latterly, and several resampling methods have been proposed in the literature. The unequal label distribution in most multilabel datasets, with disparate imbalance levels, could be a handicap while learning new classifiers. In addition, this characteristic challenges many of the existent preprocessing algorithms. Furthermore, the concurrence between imbalanced labels can make harder the learning from certain labels. These are what we call difficult labels. In this work, the problem of difficult labels is deeply analyzed, its influence in multilabel classifiers is studied, and a novel way to solve this problem is proposed. Specific metrics to assess this trait in multilabel datasets, called SCUMBLE (Score of ConcUrrence among iMBalanced LabEls) and SCUMBLELbl , are presented along with REMEDIAL (REsampling MultilabEl datasets by Decoupling highly ImbAlanced Labels), a new algorithm aimed to relax label concurrence. How to deal with this problem using the R mldr package is also outlined. [ABSTRACT FROM AUTHOR]

Abstract In this paper we propose to combine speech-based and linguistic classification in order to obtain better emotion recognition results for user spoken utterances. Usually these approaches are considered in isolation and even developed by different communities working on emotion recognition and sentiment analysis. We propose modeling the users emotional state by means of the fusion of the outputs generated with both approaches, taking into account information that is usually neglected in the individual approaches such as the interaction context and errors, and the peculiarities of transcribed spoken utterances. The fusion approach allows to employ different recognizers and can be integrated as an additional module in the architecture of a spoken conversational agent, using the information generated as an additional input for the dialog manager to decide the next system response. We have evaluated our proposal using three emotionally-colored databases and obtained very positive results. [ABSTRACT FROM AUTHOR]

Abstract Collaborative representation-based classifier (CRC) has achieved superior classification performance in the field of face recognition. However, the performance of CRC will degrade significantly when facing non-linear structural data. To address this problem, many kernel CRC (KCRC) methods have been proposed. These methods usually use a predetermined kernel function which is difficult to be selected. In addition, how to select appropriate parameters remains a challenging problem. Hence, multiple kernel technology (MKL) is applied on CRC, which called MK-CRC. However, it only considers the representation errors while ignoring the class label information in the training process. In this paper, we propose a multiple kernel locality-constrained collaborative representation-based classifier (MKLCRC) which is the multiple kernel extension of CRC and considers the local structures of data. Based on the classification rule of MKLCRC, we propose a dimensionality reduction (DR) method called multiple kernel locality-constrained collaborative representation-based discriminant projection (MKLCR-DP). The goal of MKLCR-DP is to learn a projection matrix and a set of kernel weights to generate a low-dimensional subspace where the between-class reconstruction errors are maximized and the within-class reconstruction errors are minimized. Thus MKLCRC can achieve better performance in this low-dimensional subspace. The proposed method can be efficiently optimized with the trace ratio optimization. Experiments on AR, extended Yale B, FERET, CMU PIE and LFW face databases demonstrate that our method outperforms related state-of-the-art algorithms. [ABSTRACT FROM AUTHOR]

Abstract With the development of statistical methods and machine learning algorithms, credit scoring is no longer a task merely based on experience. From single base classifiers to ensemble classifiers and hybrid models, researches have been focusing on combining classifiers and hybridizing with artificial intelligence algorithms to improve performance of the models. Ensemble classifiers have been proven to have a better predictive accuracy than single classifiers, but the method of ensemble affects performance and is worth studying. This study is based on the ensemble of five of the most widely recognized base classifiers in credit scoring, i.e. logistic regression, support vector machine, neural network, gradient boosting decision tree and random forest. It proposes a new method of selecting classifiers using Genetic Algorithm after they are trained, considering both the accuracy and diversity of the ensemble. Besides, unsupervised clustering is integrated with a fuzzy assignment procedure in the model, to make more use of the data pattern and improve performance. The proposed CF-GA-Ens model is tested on three credit scoring datasets (Australian, German, Japanese) and three performance measures (accuracy, AUC, F-score), and the results show that our classifier selection and clustering procedures have a positive impact on all performance measures. [ABSTRACT FROM AUTHOR]

The semantic segmentation task is highly related to detection and apparently can provide complementary information for detection. In this paper, we propose integrating deep semantic segmentation feature maps into the original pedestrian detection framework which combines feature channels with AdaBoost classifiers. Firstly, we develop shallow-deep channels by concatenating shallow hand-crafted and deep segmentation channels to capture appearance clues as well as semantic attributes. Then a set of manually designed filters are utilized on the new channels to generate more response feature maps. Finally a cascade AdaBoost classifier is learned for hard negatives selection and pedestrian detection. With abundant feature information, our proposed detector achieves superior results on Caltech USA 10x and ETH dataset. [ABSTRACT FROM AUTHOR]

Feature selection methods are usually evaluated by wrapping specific classifiers and datasets in the evaluation process, resulting very often in unfair comparisons between methods. In this work, we develop a theoretical framework that allows obtaining the true feature ordering of two-dimensional sequential forward feature selection methods based on mutual information, which is independent of entropy or mutual information estimation methods, classifiers, or datasets, and leads to an undoubtful comparison of the methods. Moreover, the theoretical framework unveils problems intrinsic to some methods that are otherwise difficult to detect, namely inconsistencies in the construction of the objective function used to select the candidate features, due to various types of indeterminations and to the possibility of the entropy of continuous random variables taking null and negative values. [ABSTRACT FROM AUTHOR]

Considering sentiment analysis of microblogs plays an important role in behavior analysis of social media, there has been a significant progress in this area recently. However, most researches are topic-ignored and neglect the sentimental orientation towards different topics. We propose two combined methods for topic-related Chinese message sentiment analysis. One is a graph-based ranking model of LT-IGT which takes both local and global topical information into consideration. And the other is a method of exploring sentimental features on expanded topical words with word embedding which considers both the syntactic and semantic information. These two methods are integrated into a topic-related Chinese message sentiment classifier. Experimental results on SIGHAN8 dataset show the outperformance of this approach compared with other well-known methods on sentiment analysis of topic-related Chinese message. [ABSTRACT FROM AUTHOR]

k NN is extensively studied for multilabel classification in the literature. Several k NN-based multilabel learning algorithms have been witnessed during the past years. They usually take k NN as their base classifiers to construct classification models, and then predict the class labels by virtue of Bayesian or majority rules. In this paper, a nearest neighbor selection for multilabel classification is proposed. Specifically, the target labels of new data are predicted with the help of those relevant and reliable data, which explored by the concept of shelly nearest neighbor. For effectiveness, the certainty factor is further adopted to well address the problem of unbalanced and uncertain data. The comparison experiments with eleven popular multilabel classifiers are conducted on ten benchmark data sets. The experimental results show that the performance of the proposed method is competitive and outperforms the popular multilabel classifiers in most cases. [ABSTRACT FROM AUTHOR]

Many pattern recognition problems confront two sources of annotation ambiguity where (1) multiple annotators have provided their versions of a class label which may not be consistent with one another, which forms multi-labeler learning; (2) and meanwhile a class label is associated with a bag of input vectors or instances rather than each individual instance and a bag is positive for a class label as long as one of its instances shows an evidence of that class, which is often referred to as multi-instance learning. Existing methods for multi-labeler learning and multi-instance learning only address one source of the labeling ambiguity. They are not trivially feasible to tackle the dual ambiguity problem. We hence propose a novel optimization framework by modifying the hinge loss to employ the weighted consensus of different labelers׳ labels and further generalizing the notion of loss functions to bags of multiple instances. The proposed formulation can be approximately solved by two mathematically tractable models that accommodate two types of labeling bias. An alternating optimization algorithm has been derived to efficiently solve the two models. The proposed algorithms outperform existing methods on benchmark data sets collected for document classification, real-life crowd-sourced data sets, and a medical problem of heart wall motion analysis with diagnoses from multiple radiologists. [ABSTRACT FROM AUTHOR]

Abstract: We consider the classifier ensemble problem in this paper. Due to its superior performance to individual classifiers, class ensemble has been intensively studied in the literature. Generally speaking, there are two prevalent research directions on this, i.e., to diversely generate classifier components, and to sparsely combine multiple classifiers. While most current approaches are emphasized on either sparsity or diversity only, we investigate the classifier ensemble by learning both sparsity and diversity simultaneously. We manage to formulate the classifier ensemble problem with the sparsity or/and diversity learning in a general framework. In particular, the classifier ensemble with sparsity and diversity can be represented as a mathematical optimization problem. We then propose a heuristic algorithm, capable of obtaining ensemble classifiers with consideration of both sparsity and diversity. We exploit the genetic algorithm, and optimize sparsity and diversity for classifier selection and combination heuristically and iteratively. As one major contribution, we introduce the concept of the diversity contribution ability so as to select proper classifier components and evolve classifier weights eventually. Finally, we compare our proposed novel method with other conventional classifier ensemble methods such as Bagging, least squares combination, sparsity learning, and AdaBoost, extensively on UCI benchmark data sets and the Pascal Large Scale Learning Challenge 2008 webspam data. The experimental results confirm that our approach leads to better performance in many aspects. [Copyright &y& Elsevier]

Text classification/categorization (TC) is to assign new unlabeled natural language documents to the predefined thematic categories. Centroid-based classifier (CC) has been widely used for TC because of its simplicity and efficiency. However, it has also been long criticized for its relatively low classification accuracy compared with state-of-the-art classifiers such as support vector machines (SVMs). In this paper, we find that for CC using only border instances rather than all instances to construct centroid vectors can obtain higher generalization accuracy. Along this line, we propose Border-Instance-based Iteratively Adjusted Centroid Classifier (IACC_BI), which relies on the border instances found by some routines, e.g. 1-Nearest-and-1-Furthest-Neighbors strategy, to construct centroid vectors for CC. IACC_BI then iteratively adjusts the initial centroid vectors according to the misclassified training instances. Our extensive experiments on 11 real-world text corpora demonstrate that IACC_BI improves the performance of centroid-based classifiers greatly and obtains classification accuracy competitive to the well-known SVMs, while at significantly lower computational costs. [ABSTRACT FROM AUTHOR]

Abstract: The paper shows the possibilities of generalizing the two-class classification into multi-class classification by means of a fuzzy inference system. Fuzzy combiner harnesses the support values from classifiers to provide final response having no other restrictions on their structure. We compare proposed combination methods with ECOC and two variations of decision templates, based on Euclidean and symmetric distance. The effectiveness of the proposed combination method based on the fuzzy logic theory is also evaluated via computer experiments carried out on benchmark datasets. [Copyright &y& Elsevier]