<p>Autism Spectrum Disorder (ASD) detection and classification bear huge importance in early diagnosis and intervention, which will massively raise the quality of life in autistic individuals. Most traditional approaches have difficulties in catching the sophisticated patterns of ASD-related behaviors, which often lead to suboptimal accuracy, high false-positive rates, and poor generalization across datasets. This paper introduces a new approach: the NeD-TraF framework and the Eagle Vision Arithmetic Search (EVA-Search) algorithm. This advanced deep learning architecture is very efficient in extracting ASD-salient features and further classifying them accordingly. In contrast, a dynamically optimized learning rate decay would result in much quicker convergence with the EVA-Search method. As discussed above, the methods described propose novelty at the levels of adaptive feature extraction, improved computation efficiency, and a modular approach that guarantees scalability on various datasets. It verifies the efficiency of the proposed framework with the Reddit Mental Health Dataset and ASD Screening Dataset. Experimental results reveal that the model is able to outperform six state-of-the-art methods with higher accuracy and F1-scores, such as 99%, at reduced computational time. Thus, the superiority of the current framework signifies that it is much capable of capturing the complexities in a relationship and thereby suitable for real-world applications. This provides a transformational solution in ASD detection and classification, setting a new standard in precision and efficiency in computational diagnostics.</p>