Early Autism Spectrum Disorder Detection Using SMOTE‐Enhanced 1D‐CNN and Behavioral Screening Data

Autism spectrum disorder (ASD) encompasses a range of neurodevelopmental conditions characterized by difficulties in social interaction, repetitive behaviors, speech, and nonverbal communication. Affecting approximately 1 in 36 children globally, ASD poses significant diagnostic challenges, particularly in early childhood, due to the absence of definitive medical tests. Current diagnostic practices rely heavily on behavioral observations and developmental history, often leading to delays in intervention. Early and accurate detection is essential for initiating timely support and improving developmental outcomes. In this study, we present a novel deep learning framework that integrates a one‐dimensional convolutional neural network (1D‐CNN) with the synthetic minority oversampling technique (SMOTE) to detect ASD using only 10 behavioral screening features (A1–A10). The dataset, derived from publicly available screening questionnaires, exhibited significant class imbalance, which was mitigated using SMOTE to improve model sensitivity and fairness. The proposed CNN + SMOTE model demonstrated superior performance, achieving an accuracy of 99.31%, surpassing traditional classifiers such as random forest (98.0%) and XGBoost (98.5%). Model evaluation was conducted using standard metrics including accuracy, precision, recall, F1‐score, and AUC, with validation via k‐fold cross‐validation. The results underscore the effectiveness of lightweight deep learning models augmented by oversampling techniques in developing accurate, interpretable, and scalable tools for ASD screening particularly in resource‐limited settings where access to advanced diagnostic technologies may be constrained.