Large Feature Space Research Articles

Application of a niched Pareto genetic algorithm for selecting features for nuclear transients classification

Feature selection for transient classification is the problem of choosing among several monitored parameters (i.e., the features) to be used for efficiently recognizing the developing transient patterns. It is a critical issue for the application of “on condition” diagnostic techniques in complex systems, such as the nuclear power plants, where hundreds of parameters are measured. Indeed, irrelevant and noisy features have been shown to unnecessarily increase the complexity of the classification problem and degrade the diagnostic performance. In this paper, the problem of selecting the features to be used for efficient transient classification is tackled by means of multiobjective genetic algorithms. The approach leads to the identification of a family of equivalently optimal subsets of features, in the Pareto sense. However, difficulties in the convergence of the standard Pareto-based multiobjective genetic algorithm search in large feature spaces may arise in terms of representativeness of the identified Pareto front whose elements may turn out to be unevenly distributed in the objective functions space, thus not providing a full picture of the potential Pareto-optimal solutions. To overcome this problem, a niched Pareto genetic algorithm is embraced in this work. The performance of the feature subsets examined during the search is evaluated in terms of two optimization objectives: the classification accuracy of a Fuzzy K-Nearest Neighbors classifier and the number of features in the subsets. During the genetic search, the algorithm applies a controlled “niching pressure” to spread out the population in the search space so that convergence is shared on different niches of the Pareto front, which is thus evenly covered. The method is tested on a diagnostic problem characterized by a very large number of process features available for the classification of simulated transients in the feedwater system of a boiling water reactor. The dynamics of the transient signals is captured by wavelet decomposition, which actually increases the complexity of the search for the optimal feature subsets by triplicating the number of features to be considered. © 2008 Wiley Periodicals, Inc.

Computational prediction models for cancer classification using mass spectrometry data

Classification of human complex diseases such as cancers using high-throughput mass spectrometry data generated by modern proteomic technology has quickly become an attractive topic of research in bioinformatics. However, successful applications of such proteomic strategies for early disease detection are greatly dependent on the effectiveness of computational models for data analysis. Ultimately, the extraction of appropriate features that can represent the identities of different classes plays the frontal critical factor for any difficult classification problems. In addition, another major problem associated with pattern recognition is how to effectively handle a large feature space. This paper addresses these two frontal issues for Mass Spectrometry (MS) classification. We apply two computational prediction models to extract features of MS data and then use vector quantisation to reduce the feature storage. We also introduce the technique of information fusion for classification enhancement. The proposed methodology was tested using an MS-based ovarian cancer dataset and the results were found to be superior to a support vector machine approach using a different feature for the same data.

Random subspace method for multivariate feature selection

In a growing number of domains data captured encapsulates as many features as possible. This poses a challenge to classical pattern recognition techniques, since the number of samples often still is limited with respect to the number of features. Classical pattern recognition methods suffer from the small sample size, and robust classification techniques are needed. In order to reduce the dimensionality of the feature space, the selection of informative features becomes an essential step towards the classification task. The relevance of the features can be evaluated either individually (univariate approaches), or in a multivariate manner. Univariate approaches are simple and fast, therefore appealing. However, possible correlation and dependencies between the features are not considered. Therefore, multivariate search techniques may be helpful. Several limitations restrict the use of multivariate searches. First, they are prone to overtraining, especially in p ≫ n (many features and few samples) settings. Secondly, they can be computationally too expensive when dealing with a large feature space. We introduce a new multivariate search technique, that is less sensitive to the noise in the data and computationally feasible as well. We compare our approach with several multivariate and univariate feature selection techniques, on an artificial dataset which provides us with ground truth information, and on a real dataset. The results show the importance of multivariate search techniques and the robustness and reliability of our novel multivariate feature selection method.

Feature learning with a genetic algorithm for fluorescence fingerprinting of plant species

Proper feature analysis facilitates recognition by focusing the process to those characteristics of observed data that carry the most significant information for the given classification task. In this paper we address the problem of feature selection from a different point of view. Instead of searching for a feature subset out of a large set of predefined candidate features we consider the situation where, given the form of the features and an algorithm for extracting them from the data, the optimizer tunes the feature extraction parameters to improve class separability. This process of feature learning will be solved by the means of a genetic algorithm. The optimized feature set is subsequently used in a neural network classifier. The performance of the feature learning approach is demonstrated with the problem of automatic identification of plant species from their fluorescence induction curves. The general approach should also be useful with other pattern recognition problems where a priori unknown characteristics are extracted from a large feature space.

A novel method of protein secondary structure prediction with high segment overlap measure: support vector machine approach

We have introduced a new method of protein secondary structure prediction which is based on the theory of support vector machine (SVM). SVM represents a new approach to supervised pattern classification which has been successfully applied to a wide range of pattern recognition problems, including object recognition, speaker identification, gene function prediction with microarray expression profile, etc. In these cases, the performance of SVM either matches or is significantly better than that of traditional machine learning approaches, including neural networks.The first use of the SVM approach to predict protein secondary structure is described here. Unlike the previous studies, we first constructed several binary classifiers, then assembled a tertiary classifier for three secondary structure states (helix, sheet and coil) based on these binary classifiers. The SVM method achieved a good performance of segment overlap accuracy SOV=76.2 % through sevenfold cross validation on a database of 513 non-homologous protein chains with multiple sequence alignments, which out-performs existing methods. Meanwhile three-state overall per-residue accuracy Q3 achieved 73.5 %, which is at least comparable to existing single prediction methods. Furthermore a useful “reliability index” for the predictions was developed. In addition, SVM has many attractive features, including effective avoidance of overfitting, the ability to handle large feature spaces, information condensing of the given data set, etc. The SVM method is conveniently applied to many other pattern classification tasks in biology.

Knowledge-based analysis of microarray gene expression data by using support vector machines.

We introduce a method of functionally classifying genes by using gene expression data from DNA microarray hybridization experiments. The method is based on the theory of support vector machines (SVMs). SVMs are considered a supervised computer learning method because they exploit prior knowledge of gene function to identify unknown genes of similar function from expression data. SVMs avoid several problems associated with unsupervised clustering methods, such as hierarchical clustering and self-organizing maps. SVMs have many mathematical features that make them attractive for gene expression analysis, including their flexibility in choosing a similarity function, sparseness of solution when dealing with large data sets, the ability to handle large feature spaces, and the ability to identify outliers. We test several SVMs that use different similarity metrics, as well as some other supervised learning methods, and find that the SVMs best identify sets of genes with a common function using expression data. Finally, we use SVMs to predict functional roles for uncharacterized yeast ORFs based on their expression data.