The prevalence of autism spectrum disorder (ASD) witnesses a sharp increasing in recent years, and early diagnosis and intervention of ASD are critically needed. This study explored the efficacy of Deep Neural Networks (DNN) in diagnosing ASD among children aged 0 to 10. Utilizing the latest dataset derived from the ASDTests mobile application, which encompasses behavioral characteristics of over 2,000 children, we implemented a DNN model to capture complex non-linear patterns indicative of ASD. The results of comparative analysis with traditional machine learning models revealed DNN's superior accuracy in predicting ASD, indicating that the DNN achieved a significant improvement in identifying minority classes post-imbalance learning treatment. The promising results, including the 99.55% accuracy rate, paved the way for future investigations into integrating DNN with multimodal data analysis and other advanced algorithms to enhance early diagnostic processes and intervention strategies for ASD.