Class imbalances versus small disjuncts

Emotion recognition from single-trial EEG based on kernel Fisher's emotion pattern and imbalanced quasiconformal kernel support vector machine

Electroencephalogram-based emotion recognition (EEG-ER) has received increasing attention in the fields of health care, affective computing, and brain-computer interface (BCI). However, satisfactory ER performance within a bi-dimensional and non-discrete emotional space using single-trial EEG data remains a challenging task. To address this issue, we propose a three-layer scheme for single-trial EEG-ER. In the first layer, a set of spectral powers of different EEG frequency bands are extracted from multi-channel single-trial EEG signals. In the second layer, the kernel Fisher's discriminant analysis method is applied to further extract features with better discrimination ability from the EEG spectral powers. The feature vector produced by layer 2 is called a kernel Fisher's emotion pattern (KFEP), and is sent into layer 3 for further classification where the proposed imbalanced quasiconformal kernel support vector machine (IQK-SVM) serves as the emotion classifier. The outputs of the three layer EEG-ER system include labels of emotional valence and arousal. Furthermore, to collect effective training and testing datasets for the current EEG-ER system, we also use an emotion-induction paradigm in which a set of pictures selected from the International Affective Picture System (IAPS) are employed as emotion induction stimuli. The performance of the proposed three-layer solution is compared with that of other EEG spectral power-based features and emotion classifiers. Results on 10 healthy participants indicate that the proposed KFEP feature performs better than other spectral power features, and IQK-SVM outperforms traditional SVM in terms of the EEG-ER accuracy. Our findings also show that the proposed EEG-ER scheme achieves the highest classification accuracies of valence (82.68%) and arousal (84.79%) among all testing methods.

Analysis of sampling techniques for imbalanced data: An n = 648 ADNI study

Many neuroimaging applications deal with imbalanced imaging data. For example, in Alzheimer's Disease Neuroimaging Initiative (ADNI) dataset, the mild cognitive impairment (MCI) cases eligible for the study are nearly two times the Alzheimer's disease (AD) patients for structural magnetic resonance imaging (MRI) modality and six times the control cases for proteomics modality. Constructing an accurate classifier from imbalanced data is a challenging task. Traditional classifiers that aim to maximize the overall prediction accuracy tend to classify all data into the majority class. In this paper, we study an ensemble system of feature selection and data sampling for the class imbalance problem. We systematically analyze various sampling techniques by examining the efficacy of different rates and types of undersampling, oversampling, and a combination of over and undersampling approaches. We thoroughly examine six widely used feature selection algorithms to identify significant biomarkers and thereby reduce the complexity of the data. The efficacy of the ensemble techniques is evaluated using two different classifiers including Random Forest and Support Vector Machines based on classification accuracy, area under the receiver operating characteristic curve (AUC), sensitivity, and specificity measures. Our extensive experimental results show that for various problem settings in ADNI, (1) a balanced training set obtained with K-Medoids technique based undersampling gives the best overall performance among different data sampling techniques and no sampling approach; and (2) sparse logistic regression with stability selection achieves competitive performance among various feature selection algorithms. Comprehensive experiments with various settings show that our proposed ensemble model of multiple undersampled datasets yields stable and promising results.

Design of an evolutionary approach for intrusion detection

A novel evolutionary approach is proposed for effective intrusion detection based on benchmark datasets. The proposed approach can generate a pool of noninferior individual solutions and ensemble solutions thereof. The generated ensembles can be used to detect the intrusions accurately. For intrusion detection problem, the proposed approach could consider conflicting objectives simultaneously like detection rate of each attack class, error rate, accuracy, diversity, and so forth. The proposed approach can generate a pool of noninferior solutions and ensembles thereof having optimized trade-offs values of multiple conflicting objectives. In this paper, a three-phase, approach is proposed to generate solutions to a simple chromosome design in the first phase. In the first phase, a Pareto front of noninferior individual solutions is approximated. In the second phase of the proposed approach, the entire solution set is further refined to determine effective ensemble solutions considering solution interaction. In this phase, another improved Pareto front of ensemble solutions over that of individual solutions is approximated. The ensemble solutions in improved Pareto front reported improved detection results based on benchmark datasets for intrusion detection. In the third phase, a combination method like majority voting method is used to fuse the predictions of individual solutions for determining prediction of ensemble solution. Benchmark datasets, namely, KDD cup 1999 and ISCX 2012 dataset, are used to demonstrate and validate the performance of the proposed approach for intrusion detection. The proposed approach can discover individual solutions and ensemble solutions thereof with a good support and a detection rate from benchmark datasets (in comparison with well-known ensemble methods like bagging and boosting). In addition, the proposed approach is a generalized classification approach that is applicable to the problem of any field having multiple conflicting objectives, and a dataset can be represented in the form of labelled instances in terms of its features.

The role of balanced training and testing data sets for binary classifiers in bioinformatics

Training and testing of conventional machine learning models on binary classification problems depend on the proportions of the two outcomes in the relevant data sets. This may be especially important in practical terms when real-world applications of the classifier are either highly imbalanced or occur in unknown proportions. Intuitively, it may seem sensible to train machine learning models on data similar to the target data in terms of proportions of the two binary outcomes. However, we show that this is not the case using the example of prediction of deleterious and neutral phenotypes of human missense mutations in human genome data, for which the proportion of the binary outcome is unknown. Our results indicate that using balanced training data (50% neutral and 50% deleterious) results in the highest balanced accuracy (the average of True Positive Rate and True Negative Rate), Matthews correlation coefficient, and area under ROC curves, no matter what the proportions of the two phenotypes are in the testing data. Besides balancing the data by undersampling the majority class, other techniques in machine learning include oversampling the minority class, interpolating minority-class data points and various penalties for misclassifying the minority class. However, these techniques are not commonly used in either the missense phenotype prediction problem or in the prediction of disordered residues in proteins, where the imbalance problem is substantial. The appropriate approach depends on the amount of available data and the specific problem at hand.

Semi-supervised prediction of SH2-peptide interactions from imbalanced high-throughput data

Src homology 2 (SH2) domains are the largest family of the peptide-recognition modules (PRMs) that bind to phosphotyrosine containing peptides. Knowledge about binding partners of SH2-domains is key for a deeper understanding of different cellular processes. Given the high binding specificity of SH2, in-silico ligand peptide prediction is of great interest. Currently however, only a few approaches have been published for the prediction of SH2-peptide interactions. Their main shortcomings range from limited coverage, to restrictive modeling assumptions (they are mainly based on position specific scoring matrices and do not take into consideration complex amino acids inter-dependencies) and high computational complexity. We propose a simple yet effective machine learning approach for a large set of known human SH2 domains. We used comprehensive data from micro-array and peptide-array experiments on 51 human SH2 domains. In order to deal with the high data imbalance problem and the high signal-to-noise ration, we casted the problem in a semi-supervised setting. We report competitive predictive performance w.r.t. state-of-the-art. Specifically we obtain 0.83 AUC ROC and 0.93 AUC PR in comparison to 0.71 AUC ROC and 0.87 AUC PR previously achieved by the position specific scoring matrices (PSSMs) based SMALI approach. Our work provides three main contributions. First, we showed that better models can be obtained when the information on the non-interacting peptides (negative examples) is also used. Second, we improve performance when considering high order correlations between the ligand positions employing regularization techniques to effectively avoid overfitting issues. Third, we developed an approach to tackle the data imbalance problem using a semi-supervised strategy. Finally, we performed a genome-wide prediction of human SH2-peptide binding, uncovering several findings of biological relevance. We make our models and genome-wide predictions, for all the 51 SH2-domains, freely available to the scientific community under the following URLs: http://www.bioinf.uni-freiburg.de/Software/SH2PepInt/SH2PepInt.tar.gz and http://www.bioinf.uni-freiburg.de/Software/SH2PepInt/Genome-wide-predictions.tar.gz, respectively.

Identifying the medical practice after total hip arthroplasty using an integrated hybrid approach

A critical option of total hip arthroplasty (THA) is considered only when tried more conservative treatments but continued to have pain, stiffness, or problems with the function of ones hip. THA plays one of major concerns under the waves of the rapid growth of aging populations and the constrained health care resources in Taiwan. Moreover, prior studies indicated that imbalanced class distribution problems do exist in the constructed classification model and cause seriously negative effects on model performances in the health care industry. Therefore, this study proposes an integrated hybrid approach to provide an alternate method for classifying the quality (e.g., the staying length in hospital) of medical practice with an imbalanced class problem after performing a THA procedure for hip replacement patients and their doctors in the health care industry. The proposed approach is constituted by seven components: expert knowledge, global discretization, imbalanced bootstrap technique, reduct and core methods, rough sets, rule induction, and rule filter. The proposed approach is illustrated in practice by examining an experimental dataset from the National Health Insurance Research Database (NHIRD) in Taiwan. The experimental results reveal that the proposed approach has better performance than the listed methods under evaluation criteria. The output created by the rough set LEM2 algorithm is a comprehensible decision rule set that can be applied in knowledge-based health care services as desired. The analytical results provide useful THA information for both academics and practitioners and these results could be applicable to other diseases or to other countries with similar social and cultural practices.

Predicting protein-ATP binding sites from primary sequence through fusing bi-profile sampling of multi-view features

Background

Adenosine-5′-triphosphate (ATP) is one of multifunctional nucleotides and plays an important role in cell biology as a coenzyme interacting with proteins. Revealing the binding sites between protein and ATP is significantly important to understand the functionality of the proteins and the mechanisms of protein-ATP complex.

Results

In this paper, we propose a novel framework for predicting the proteins’ functional residues, through which they can bind with ATP molecules. The new prediction protocol is achieved by combination of sequence evolutional information and bi-profile sampling of multi-view sequential features and the sequence derived structural features. The hypothesis for this strategy is single-view feature can only represent partial target’s knowledge and multiple sources of descriptors can be complementary.

Conclusions

Prediction performances evaluated by both 5-fold and leave-one-out jackknife cross-validation tests on two benchmark datasets consisting of 168 and 227 non-homologous ATP binding proteins respectively demonstrate the efficacy of the proposed protocol. Our experimental results also reveal that the residue structural characteristics of real protein-ATP binding sites are significant different from those normal ones, for example the binding residues do not show high solvent accessibility propensities, and the bindings prefer to occur at the conjoint points between different secondary structure segments. Furthermore, results also show that performance is affected by the imbalanced training datasets by testing multiple ratios between positive and negative samples in the experiments. Increasing the dataset scale is also demonstrated useful for improving the prediction performances.

Using ensemble methods to deal with imbalanced data in predicting protein-protein interactions

In proteins, the number of interacting pairs is usually much smaller than the number of non-interacting ones. So the imbalanced data problem will arise in the field of protein-protein interactions (PPIs) prediction. In this article, we introduce two ensemble methods to solve the imbalanced data problem. These ensemble methods combine the based-cluster under-sampling technique and the fusion classifiers. And then we evaluate the ensemble methods using a dataset from Database of Interacting Proteins (DIP) with 10-fold cross validation. All the prediction models achieve area under the receiver operating characteristic curve (AUC) value about 95%. Our results show that the ensemble classifiers are quite effective in predicting PPIs; we also gain some valuable conclusions on the performance of ensemble methods for PPIs in imbalanced data. The prediction software and all dataset employed in the work can be obtained for free at http://cic.scu.edu.cn/bioinformatics/Ensemble_PPIs/index.html.

SYMBOLIC ONE-CLASS LEARNING FROM IMBALANCED DATASETS: APPLICATION IN MEDICAL DIAGNOSIS

When working with real-world applications we often find imbalanced datasets, those for which there exists a majority class with normal data and a minority class with abnormal or important data. In this work, we make an overview of the class imbalance problem; we review consequences, possible causes and existing strategies to cope with the inconveniences associated to this problem. As an effort to contribute to the solution of this problem, we propose a new rule induction algorithm named Rule Extraction for MEdical Diagnosis (REMED), as a symbolic one-class learning approach. For the evaluation of the proposed method, we use different medical diagnosis datasets taking into account quantitative metrics, comprehensibility, and reliability. We performed a comparison of REMED versus C4.5 and RIPPER combined with over-sampling and cost-sensitive strategies. This empirical analysis of the REMED algorithm showed it to be quantitatively competitive with C4.5 and RIPPER in terms of the area under the Receiver Operating Characteristic curve (AUC) and the geometric mean, but overcame them in terms of comprehensibility and reliability. Results of our experiments show that REMED generated rules systems with a larger degree of abstraction and patterns closer to well-known abnormal values associated to each considered medical dataset.

Automatic defect detection for TFT-LCD array process using quasiconformal kernel support vector data description

Defect detection has been considered an efficient way to increase the yield rate of panels in thin film transistor liquid crystal display (TFT-LCD) manufacturing. In this study we focus on the array process since it is the first and key process in TFT-LCD manufacturing. Various defects occur in the array process, and some of them could cause great damage to the LCD panels. Thus, how to design a method that can robustly detect defects from the images captured from the surface of LCD panels has become crucial. Previously, support vector data description (SVDD) has been successfully applied to LCD defect detection. However, its generalization performance is limited. In this paper, we propose a novel one-class machine learning method, called quasiconformal kernel SVDD (QK-SVDD) to address this issue. The QK-SVDD can significantly improve generalization performance of the traditional SVDD by introducing the quasiconformal transformation into a predefined kernel. Experimental results, carried out on real LCD images provided by an LCD manufacturer in Taiwan, indicate that the proposed QK-SVDD not only obtains a high defect detection rate of 96%, but also greatly improves generalization performance of SVDD. The improvement has shown to be over 30%. In addition, results also show that the QK-SVDD defect detector is able to accomplish the task of defect detection on an LCD image within 60 ms.

RAMOBoost: Ranked Minority Oversampling in Boosting

In recent years, learning from imbalanced data has attracted growing attention from both academia and industry due to the explosive growth of applications that use and produce imbalanced data. However, because of the complex characteristics of imbalanced data, many real-world solutions struggle to provide robust efficiency in learning-based applications. In an effort to address this problem, this paper presents Ranked Minority Oversampling in Boosting (RAMOBoost), which is a RAMO technique based on the idea of adaptive synthetic data generation in an ensemble learning system. Briefly, RAMOBoost adaptively ranks minority class instances at each learning iteration according to a sampling probability distribution that is based on the underlying data distribution, and can adaptively shift the decision boundary toward difficult-to-learn minority and majority class instances by using a hypothesis assessment procedure. Simulation analysis on 19 real-world datasets assessed over various metrics-including overall accuracy, precision, recall, F-measure, G-mean, and receiver operation characteristic analysis-is used to illustrate the effectiveness of this method.