Open AccessIdentification of Best Fit Learning Models Based on Calibration for Better Classification of Autism
Author(s) -
B. S. Roopa,
R. Manjunatha Prasad
Publication year - 2020
Publication title -
international journal of engineering and advanced technology
Language(s) - English
Resource type - Journals
ISSN - 2249-8958
DOI - 10.35940/ijeat.c5205.029320
Subject(s) - autism , computer science , artificial intelligence , machine learning , brier score , calibration , probabilistic logic , measure (data warehouse) , identification (biology) , support vector machine , statistical model , autism spectrum disorder , set (abstract data type) , a priori and a posteriori , data mining , statistics , mathematics , psychology , developmental psychology , botany , biology , programming language , philosophy , epistemology
This paper is intended to exhibit the novel approach to improve the efficiency of the supervised learning models towards the accuracy of the predictions made to classify the autism from that of the normal subject. The state of the art is about 60-75% of Autism classification accuracy. The early prediction of autism plays a vital role as the rise of autism is alarming. The invasive way to analyze the problem at the earliest would render much support to the Autism Spectrum Disorder (ASD) community. In this work, various supervised learning models are first tested on 1101 subjects with 530 ASD subjects and 571 Normal subjects. The Datasets worked are collected from Autism Brain Imaging Data Exchange (ABIDE) repository. The performance measure is calibrated in terms of Brier score which is an accuracy measure of the predictions in probabilistic way. After assessing in probabilistic way, the statistically emphasized models are then evaluated for the same set of data to validate the prediction model efficiency with their statistical measures made and hence developing the confidence of the model selection for better classification based on probability calibration (CAL). The performance evaluation of the model is tested with probability calibrated assessment and found that for given dataset the SVM and Logistic Regression provided better accuracy measure compared to other considered learning models. It is necessary to frame a hypothesis measure on the dataset before any model is deployed. This approach helps to identify the desired and validated supervised model for the given data samples.