AI-Driven Solutions for Autism Spectrum Disorder Assessment: A Robust and Reliable Deep Learning Model for Diagnosis and Prediction

Among the most burdensome neurodevelopmental conditions, Autism Spectrum Disorder affects millions of families worldwide. Early identification, in that respect, is considered a critical variable that surely enhances the effectiveness of interventions and treatments. Diagnosis of ASD is very difficult owing to the diverse symptoms of ASD; an overall thorough and meticulous assessment is required. Thus, traditional diagnostic techniques based on subjective evaluation and behavioral observations are inconsistent and delayed. Besides, most machine learning models for predictions, though promising, are many times not interpretable, hence not easy to trust or explain the clinicians in decision-making when it comes to such predictions. This paper tries to address these by proposing certain novelties in ASD detection based on deep learning, including a stack of RNNs with LSTMs. The proposed RNN+LSTM-based model learns the temporal dependencies and sequential patterns of ASD-related data, including behavioral assessment and diagnostic tests; thus, this model is particularly fitted to these. Besides enhancing accuracy and interpretability in diagnosis, it will also provide insight into drivers of detection. It overcomes the drawback in the normal machine learning approach by integrating RNN with LSTMs for more robust predictions of ASD.