Embedded Feature Importance with Threshold-based Selection for Optimal Feature Subset in Autism Screening

Abstract

The early detection of autism spectrum disorders (ASD) in children poses significant challenges due to the dynamic and progressive nature of the symptoms. To The current screening process involves a lengthy and costly series of questions covering various aspects of a child's development. To address this issue, we adopt the embedded feature selection method based on random forest and threshold-based to produce a simplified version questionnaire for Autism screening. The aim of this paper is to identify the most crucial and effective features from the Quantitative Checklist for Autism in Toddlers (Q-CHAT) by combining the strengths of threshold filtering and embedded random forest feature importance. This integration allows us to significantly reduce the number of screening questions while maintaining reliable and accurate results. Our proposed method yields a streamlined alternative to traditional screening, extracting just eight key features that achieves an impressive 96% accuracy performance. This promising approach holds the potential to revolutionize early detection and intervention programs for toddlers with autism, ultimately leading to improved outcomes.