Investigation of Statistical Feature Selection Techniques for Autism Classification Using EEG Signals

S. Thirumal and J. Thangakumar

A neurodevelopmental condition displaying a socio-communicative impairment along with interests and activities that are stereotyped is known as Autism. This is associated with hereditary to a high level. The Electroencephalogram (EEG) is a technique that is non-invasive and commonly used for acquiring an electrical impulse that is produced from the activation of neurons in the human brain. The EEG electrodes that are fixed to certain regions in the scalp in accordance with the type of study which is conducted. An automatic analysis of the EEG signals includes certain steps in signal processing and classification. The extraction of features is a process that obtains information from the signals of EEG that represent larger datasets for classification. The Wavelet Transform (WT) can be called another method of non-stationary time-scale analysis which is well-suited and can be used on the EEG signals for extracting features. These features that are extracted will be used for training a classification of the K-Nearest Neighbour (KNN). In machine learning the process of feature selection is important step in preprocessing for very effective the reduction of dimensionality, removal of irrelevant data, improving the accuracy of learning and the comprehensibility of results. A Correlation-based Feature Selection (CFS), Minimum Redundancy Maximum Relevance (MRMR) and the Information Gain (IG) have been evaluated in this work. The results of the experiment have shown that the feature selection methods help achieve better performance of the classifier.

Volume 12 | 05-Special Issue

Pages: 1254-1263

DOI: 10.5373/JARDCS/V12SP5/20201883