Classification of autism features in electroencephalography recordings using random forest method

Abstract

Autism Spectrum Disorder (ASD) is a developmental disorder that significantly impacts communication, social interaction, and behavior in children, often leading to withdrawal, repetitive behaviors, and difficulties with eye contact. Traditional diagnostic methods primarily relied on behavioral assessments, which have proven insufficient in accuracy. This study aims to enhance ASD diagnosis by employing Electroencephalography (EEG) as an objective marker to differentiate between individuals with ASD and neurotypical individuals. Utilizing a dataset from King Abdulaziz University comprising 16 children—4 neurotypical and 12 with ASD—this research implements preprocessing techniques such as Independent Component Analysis (ICA) to eliminate noise and artifacts from EEG signals. Following this, Wavelet Packet Decomposition (WPD) is applied at three levels to improve signal resolution. Statistical features including mean, variance, skewness, and kurtosis are extracted for classification purposes. The Random Forest (RF) method is then employed for classification, achieving an accuracy of 76.8%. However, classification errors predominantly arise from the imbalance in the dataset, with more data available for ASD subjects compared to neurotypical subjects. The findings reveal significant differences in statistical features between the two groups, indicating the potential of EEG technology and computational algorithms in developing a more accurate and objective ASD diagnosis system. This research contributes valuable insights for early intervention strategies and future studies aimed at improving diagnostic methodologies for children with autism.