Publication

Abstract : ASD is a serious neural illness which creates problems in social interaction, connection and other behaviors. Since there is no specific medical test for ASD, it is primarily treated through behavioural observation and child's medical history. Hence, earlier treatment is important, but it is often delayed due to parental reluctance or lack of awareness. Children with ASD simultaneously experience sensory sensitivities which can further complicate their behavioural patterns. Although screening tests exist, most of them are expensive and time-consuming. To overcome the issue of traditional methods, this study shows different ML models for classification and regression tasks, to figure out the presence of autism in children. Based on performance metrics, these models are analysed to identify the most effective method for exact prediction process. XGBoost in regression and Gradient Boosting in classification are chosen as the best models. Since the dataset's target variable is categorical, hybrid models are also introduced for categorizing the autism affected children and among various hybrid methods, the combination of XGBoost, CatBoost and Light Gradient Boosting Machine (LGBM) performed well with highly effective results.