Analysis of Randomization-Based Approaches for Autism Spectrum Disorder

Abstract

AbstractAutism spectrum disorder (ASD) is a severe neurodevelopmental disorder that affects an individual’s sensory activity, social interaction, and cognitive abilities. In the mental illnesses ASD disorder, the problem starts in infancy and affects more habits as the age progresses, progressing to adolescence and adulthood, also known as a behavioural disorder. Everything nowadays is moving towards automated software, which is relevant not just in terms of time efficiency but also in terms of cost-effectiveness. As a result, there is a pressing need in the healthcare sector to incorporate machine learning to reap the greatest benefits. Over the last two decades, randomization-based approaches such as extreme learning machines (ELM) and random vector functional link (RVFL) have gained popularity amongst researchers due to their better generalization performance. In this work, the classification capability of the ELM and RVFL models with different activation functions is investigated to estimate the autism spectrum disorder in a human population of children, adolescents, and adults on the grounds of publicly accessible UCI datasets. The attainment of the randomization-based approaches is determined using various quality measures such as accuracy, precision, recall, negative predictive value, rate of misclassification, F1-measure, G-mean, and Matthew’s correlation coefficient. From the numerical experiment results, one can show that the generalization capability of RVFL using hardlim, hyperbolic tangent function, and sigmoidal activation function is superior to ELM based on several quality measures.KeywordsExtreme learning machineRandom vector function linkASD disorderAutism traitsRandomization-based approaches