A Multimodal AI Framework for Early Autism Identification Using CbRNN–Ensemble Modelling and Advanced NCIAPPF‐Based Temporal Encoding

To improve child developmental results and enable prompt interventions, early identification of autism spectrum disorder (ASD) is crucial. Conventional diagnostic instruments, while clinically standardized, remain constrained by subjectivity, limited scalability and dependence on expert evaluation. This study introduces AutiScan, an AI-driven framework that integrates a Convolution-based Recurrent Neural Network Ensemble Model (CbRNN-EM) with the neural context-integrated adaptive reprocessing and pattern filtering (NCIAPPF) technique for capturing temporal dynamics of neural encoding. The proposed system analyses multimodal data including facial expressions, eye-gaze patterns, speech prosody and behavioural cues to extract spatial, emotional and sequential features relevant to early ASD markers. NCIAPPF enhances preprocessing by adaptively filtering noise, normalizing multimodal signals and preserving fine-grained temporal dependencies essential for robust pattern recognition. The fused representations are classified using an ensemble layer, achieving superior accuracy compared with traditional deep learning and machine learning prototypes. The results of the experiments show that AutiScan offers a dependable, scalable and comprehensible early ASD screening tool, underscoring the opportunity for practical clinical and educational implementation.