Pedro J. García-Laencina

Pattern classification with missing data: a review

Pattern classification has been successfully applied in many problem domains, such as biometric recognition, document classification or medical diagnosis. Missing or unknown data are a common drawback that pattern recognition techniques need to deal with when solving real-life classification tasks. Machine learning approaches and methods imported from statistical learning theory have been most intensively studied and used in this subject. The aim of this work is to analyze the missing data problem in pattern classification tasks, and to summarize and compare some of the well-known methods used for handling missing values.

K nearest neighbours with mutual information for simultaneous classification and missing data imputation

Missing data is a common drawback in many real-life pattern classification scenarios. One of the most popular solutions is missing data imputation by the K nearest neighbours (KNN) algorithm. In this article, we propose a novel KNN imputation procedure using a feature-weighted distance metric based on mutual information (MI). This method provides a missing data estimation aimed at solving the classification task, i.e., it provides an imputed dataset which is directed toward improving the classification performance. The MI-based distance metric is also used to implement an effective KNN classifier. Experimental results on both artificial and real classification datasets are provided to illustrate the efficiency and the robustness of the proposed algorithm.

EFFICIENT AUTOMATIC SELECTION AND COMBINATION OF EEG FEATURES IN LEAST SQUARES CLASSIFIERS FOR MOTOR IMAGERY BRAIN–COMPUTER INTERFACES

Discriminative features have to be properly extracted and selected from the electroencephalographic (EEG) signals of each specific subject in order to achieve an adaptive brain-computer interface (BCI) system. This work presents an efficient wrapper-based methodology for feature selection and least squares discrimination of high-dimensional EEG data with low computational complexity. Features are computed in different time segments using three widely used methods for motor imagery tasks and, then, they are concatenated or averaged in order to take into account the time course variability of the EEG signals. Once EEG features have been extracted, proposed framework comprises two stages. The first stage entails feature ranking and, in this work, two different procedures have been considered, the least angle regression (LARS) and the Wilcoxon rank sum test, to compare the performance of each one. The second stage selects the most relevant features using an efficient leave-one-out (LOO) estimation based on the Allens PRESS statistic. Experimental comparisons with the state-of-the-art BCI methods shows that this approach gives better results than current state-of-the-art approaches in terms of recognition rates and computational requirements and, also with respect to the first ranking stage, it is confirmed that the LARS algorithm provides better results than the Wilcoxon rank sum test for these experiments.

Efficient feature selection and linear discrimination of EEG signals

Abstract Brain–Computer Interface systems (BCIs) based on Electroencephalogram (EEG) signal processing allow us to translate the subjects brain activities into control commands for computer devices. This paper presents an efficient embedded approach for feature selection and linear discrimination of EEG signals. In the first stage, four well-known feature extraction methods are used: Power spectral features, Hjorth parameters, Autoregressive modelling and Wavelet transform. From all the obtained features, the proposed method efficiently selects and combines the most useful features for classification with less computational requirements. Least Angle Regression (LARS) is used for properly ranking each feature and, then, an efficient Leave-One-Out (LOO) estimation based on the PRESS statistic is used to choose the most relevant features. Experimental results on motor-imagery BCIs problems are provided to illustrate the competitive performance of the proposed approach against other conventional methods.

Missing data imputation on the 5-year survival prediction of breast cancer patients with unknown discrete values

Breast cancer is the most frequently diagnosed cancer in women. Using historical patient information stored in clinical datasets, data mining and machine learning approaches can be applied to predict the survival of breast cancer patients. A common drawback is the absence of information, i.e., missing data, in certain clinical trials. However, most standard prediction methods are not able to handle incomplete samples and, then, missing data imputation is a widely applied approach for solving this inconvenience. Therefore, and taking into account the characteristics of each breast cancer dataset, it is required to perform a detailed analysis to determine the most appropriate imputation and prediction methods in each clinical environment. This research work analyzes a real breast cancer dataset from Institute Portuguese of Oncology of Porto with a high percentage of unknown categorical information (most clinical data of the patients are incomplete), which is a challenge in terms of complexity. Four scenarios are evaluated: (I) 5-year survival prediction without imputation and 5-year survival prediction from cleaned dataset with (II) Mode imputation, (III) Expectation-Maximization imputation and (IV) K-Nearest Neighbors imputation. Prediction models for breast cancer survivability are constructed using four different methods: K-Nearest Neighbors, Classification Trees, Logistic Regression and Support Vector Machines. Experiments are performed in a nested ten-fold cross-validation procedure and, according to the obtained results, the best results are provided by the K-Nearest Neighbors algorithm: more than 81% of accuracy and more than 0.78 of area under the Receiver Operator Characteristic curve, which constitutes very good results in this complex scenario.

Neural networks letter: Neural architecture design based on extreme learning machine

Selection of the optimal neural architecture to solve a pattern classification problem entails to choose the relevant input units, the number of hidden neurons and its corresponding interconnection weights. This problem has been widely studied in many research works but their solutions usually involve excessive computational cost in most of the problems and they do not provide a unique solution. This paper proposes a new technique to efficiently design the MultiLayer Perceptron (MLP) architecture for classification using the Extreme Learning Machine (ELM) algorithm. The proposed method provides a high generalization capability and a unique solution for the architecture design. Moreover, the selected final network only retains those input connections that are relevant for the classification task. Experimental results show these advantages.

Classifying patterns with missing values using Multi-Task Learning perceptrons

Datasets with missing values are frequent in real-world classification problems. It seems obvious that imputation of missing values can be considered as a series of secondary tasks, while classification is the main purpose of any machine dealing with these datasets. Consequently, Multi-Task Learning (MTL) schemes offer an interesting alternative approach to solve missing data problems. In this paper, we propose an MTL-based method for training and operating a modified Multi-Layer Perceptron (MLP) architecture to work in incomplete data contexts. The proposed approach achieves a balance between both classification and imputation by exploiting the advantages of MTL. Extensive experimental comparisons with well-known imputation algorithms show that this approach provides excellent results. The method is never worse than the traditional algorithms - an important robustness property - and, also, it clearly outperforms them in several problems.

Exploring dimensionality reduction of EEG features in motor imagery task classification

A Brain-Computer Interface (BCI) system based on motor imagery (MI) identifies patterns of electrical brain activity to predict the user intention while certain movement imagination tasks are performed. Currently, one of the most important challenges is the adaptive design of a BCI system. For solving it, this work explores dimensionality reduction techniques: once features have been extracted from Electroencephalogram (EEG) signals, the high-dimensional EEG data has to be mapped onto a new reduced feature space to make easier the classification stage. Besides the standard sequential feature selection methods, this paper analyzes two unsupervised transformation-based approaches – Principal Component Analysis and Locality Preserving Projections – and the Local Fisher Discriminant Analysis (LFDA), which works in a supervised manner. The dimensionality in the projected space is chosen following a wrapper-based approach by an efficient leave-one-out estimation. Experiments have been conducted on five novice subjects during their first sessions with MI-based BCI systems in order to show that the appropriate use of dimensionality reduction methods allows increasing the performance. In particular, obtained results show that LFDA gives a significant enhancement in classification terms without increasing the computational complexity and, then, it is a promising technique for designing MI-based BCI system.

Automatic and Adaptive Classification of Electroencephalographic Signals for Brain Computer Interfaces

Extracting knowledge from electroencephalographic (EEG) signals has become an increasingly important research area in biomedical engineering. In addition to its clinical diagnostic purposes, in recent years there have been many efforts to develop brain computer interface (BCI) systems, which allow users to control external devices only by using their brain activity. Once the EEG signals have been acquired, it is necessary to use appropriate feature extraction and classification methods adapted to the user in order to improve the performance of the BCI system and, also, to make its design stage easier. This work introduces a novel fast adaptive BCI system for automatic feature extraction and classification of EEG signals. The proposed system efficiently combines several well-known feature extraction procedures and automatically chooses the most useful features for performing the classification task. Three different feature extraction techniques are applied: power spectral density, Hjorth parameters and autoregressive modelling. The most relevant features for linear discrimination are selected using a fast and robust wrapper methodology. The proposed method is evaluated using EEG signals from nine subjects during motor imagery tasks. Obtained experimental results show its advantages over the state-of-the-art methods, especially in terms of classification accuracy and computational cost.

Pattern Classification with Missing Values using Multitask Learning

In many real-life applications it is important to know how to deal with missing data (incomplete feature vectors). The ability of handling missing data has become a fundamental requirement for pattern classification because inappropriate treatment of missing data may cause large errors or false results on classification. A novel effective neural network is proposed to handle missing values in incomplete patterns with multitask learning (MTL). In our approach, a MTL neural network learns in parallel the classification task and the different tasks associated to incomplete features. During the MTL process, missing values are estimated or imputed. Missing data imputation is guided and oriented by the classification task, i.e., imputed values are those that contribute to improve the learning. We prove the robustness of this MTL neural network for handling missing values in classification problems from UCI database.