An Improvised Multilayer Perceptron Network Using Boosted Regression Tree Based Missing Value Imputation And Fuzzy Backward Elimination Feature Selection For Autism Disease Prediction

Abstract

Autism is also known as behavioral disease which affects the communication skill and social interaction with others. This neuro-syndrome begins at a person’s childhood and continues throughout their life. But early detection of autism may positively assist the victim to maintain both their physical and mental health. Emergence of machine learning in the field of medical research greatly improves the diagnosis of disease at their earlier stages. Thus, this paper focuses on developing a improvised classification model to increase the accurate classification by enhancing the quality of autism dataset. Though, there many existing research works are there to classify the autism, importance on quality of dataset is not considered. In this proposed work the raw autism dataset collected from Kaggle repository with missing values is well treated by adapting two main preprocessing method. In this work boosted regression tree is used for imputing the missing values of autism to complete dataset. Fuzzy backward feature elimination is applied for reducing the feature set involved in classification and it also influence the accuracy by maximizing the relevancy of attributes and minimizing the redundancy among them. The Multilayer perceptron is used as classifier for handling the imbalance class issue in autism detection. From the results obtained the efficacy of the proposed improvised multilayer perceptron is proved more prominently in autism detection.