LAGRANGIAN MATCHED FILTER WITH REGRESSIVE PERCETRON DEEP CLASSIFIER FOR AUTISM DISORDER DETECTION WITH EEG SIGNALS

Abstract

People through ASD frequently contain cognitive disabilities. Efficient association among distinct brain regions is indispensable for ordinary cognition. Electro Encephalo Graphy (EEG) has extensively utilized at neurological disease recognition. Preceding studies on identifying ASD through EEG information centered on time domain, channel and spatial associated aspects. Mainly these studies not preserved the essential information while performing dimensionality reduction. However, such approaches may result in compromising accuracy and increasing training time. To resolve this issue, Deep Belief Network Classification technique for identifying ASD through EEG information using Lagrangian Matched Filter with Truncated Regressive Piecewise Perceptron (LMF-TRPP). The proposed LMF-TRPP method is split asone input layer, three hidden layers and one output layer for carry out autism spectrum detection with EEG signals. Initially, the samples gathered from EEG signals database are provided as input via input layer. To eliminate noise (such as muscle activity in the head, blinking) from the raw input EEG signals, Lagrangian Matched Filter denoising model is carried out in the first hidden layer. Following which relevant features are selected in the second hidden layer by applying Truncated Tobit Regressive Feature Selection model. By selecting this feature selection model aids in minimizing dimensionality while preserving the essential information for accurate classification. Finally Perceptron Piecewise Classification process is modeled in the third hidden layer for performing efficient classification as autism EEG signal or non-autism EEG signal with selected relevant features. Experimental outcomes demonstrated LMF-TRPP technique attained better result compared through two ASD detection techniques that has favorable potentiality to impart an accurate diagnosis to assist clinicians. In conclusion, this work highlights the effectiveness of Deep Belief Network Classification in EEG signal processing, offering valuable contributions with improved precision and computationally efficient for additional precise neurological disease classification as well as diagnosis.