Feature Selection Using Various Hybrid Algorithms for Speech Recognition

Abstract

Feature selection aims at selection of relevant features by applying certain selection criteria leading to high classification performance. Many practical problems consist of number of features, which makes classification as challenging task. With all these features system load and computation time will increase. Various feature selection and ranking methods are available to eliminate irrelevant features.In this paper the dataset used is available in analog form, which is converted into digital form for feature extraction and the obtained dataset is used to perform the experiments. Various algorithms like Minimum Redundancy Maximum Relevance (MRMR), Correlation based Feature Selection (CFS), t-Test, Chi Square, Fast Correlation Based Feature selection (FCBF), Fisher Score, Gini Index, Information Gain, Krushkal Wallis and BlogReg are used to rank and select features from speech dataset. These algorithms are combined with genetic algorithm; this hybrid approach minimizes the selected features. SVM is used for classification and prediction of data for selected features. The classification accuracy of dataset with reduced features obtained using above mentioned algorithms and hybrid approach is compared.