Enhancing autism detection through machine learning models focusing on behavioral analysis

Abstract

Autism Spectrum Disorder (ASD) is a complex and enduring condition characterized by challenges related to communication and behavior. It is a complex neurological disorder related to an individual’s psychological difficulties, which eventually impact their behavior or reactions to the outside world. While it is feasible to detect autism symptoms at any stage of an individual’s life, there is a greater likelihood of detection within the first two years after birth, as differences in typical activities, communication gaps, or a lack of understanding typically become more noticeable during this early developmental period. The paper suggests a deep learning-based method that makes use of behavior to identify autism in both adults and children by analyzing their behavioral characteristics through machine learning approaches and determining a process that makes autism detection easier and cost-effective. The recommended approach works by behavioral monitoring of children and adult datasets that were collected from online platforms and went through successive processing and finally, those datasets were applied to different models with the help of binary classification towards the end to determine autism detection correctly. Behavioral data includes a range of indicators, including patterns of social interaction, communication abilities, and recurrent actions. For behavior analysis, we implemented specific models like KNN, Random Forest, CatBoost, SVM, GradientBoost, and Logistic Regression and also ensembled models by incorporating a few of our pre-trained models together to give better accuracy rates. We have also integrated different confusion matrices in our paper. This will help in evaluating and fine-tuning the performance of our detection model. We have acquired behavioral datasets from publicly available platforms called UC Irvine Machine Learning Repository and Kaggle. Our primary goal is to improve the accuracy of autism detection or contribute to research by developing a comprehensive research paper. This paper aims to facilitate model comparisons and streamline the autism detection process using advanced machine learning techniques available today.