Nonclinical Approaches to Autism Trait Prediction Using Deep Learning: An Extensive Review

Abstract

Autism spectrum disorder (ASD) is a neurodevelopmental condition characterized by difficulties in communication and social behavior. Early identification of autistic traits is necessary for timely treatment. However, clinical diagnostic methods remain costly, time-consuming, and inaccessible. AI-guided nonclinical screening techniques such as eye tracking and behavior analysis may offer a promising alternative. This is the first systematic bibliometric analysis attempting to discuss the effectiveness, limitations, and trends of AI-based methods proposed for identifying autistic traits using nonclinical analysis. The current study is an unprecedented bibliometric analysis of 152 Scopus and Web of Science articles (2018–2025). This analysis uses VOSviewer and Gephi to map different relations. Deep learning and hybrid transfer learning yield better results. However, most of the proposed methods possess low specificity. The proposed research establishes eye-gaze analysis as the most commonly employed technique in the present study, while key behavioral cues—facial expressions, attention changes, and verbal patterns—are yet to be studied. The key shortcoming is that no standardized evaluation system exists, with earlier work lacking rigorous benchmarks for specificity, interpretability, and bias mitigation. Furthermore, while transfer learning techniques are widely used due to dataset scarcity, no publicly available video-based datasets exist, restricting the development of multimodal ASD screening models. To fill these gaps, this paper suggests a benchmarking framework that focuses on multifeature evaluation (facial expressions and attention shifts) for better diagnosis. In addition, standardized specificity thresholds support clinical reliability and foster geographically diverse, openly available datasets to obtain fairness and generalizability. This study provides the bibliometric synthesis of nonclinical AI-based ASD screening. It proposes a benchmarking framework with multifeature evaluation, standardized specificity thresholds, and diverse open datasets and maps data modalities to suitable deep learning models. These contributions provide practical guidance and deliver actionable insights to advance scalable, multimodal, and interpretable ASD screening tools for early, noninvasive detection.

Received: 20 May 2025 | Revised: 10 September 2025 | Accepted: 15 October 2025

Conflicts of Interest

The authors declare that they have no conflicts of interest to this work.

Data Availability Statement

The data that support the findings of this study are openly available in Kaggle at https://www.kaggle.com/datasets/shahriarrafi1071/autism-dataset?select=AutismDataset.

Author Contribution Statement

Ranjeet Vasant Bidwe: Conceptualization, Methodology, Software, Validation, Formal analysis, Investigation, Resources, Data curation, Writing – original draft, Writing – review & editing, Visualization. Sashikala Mishra: Conceptualization, Methodology, Software, Validation, Data curation, Writing – review & editing, Supervision, Project administration. Simi Bajaj: Conceptualization, Methodology, Software, Validation, Writing – review & editing, Supervision, Project administration. Suraj Sawant: Validation, Formal analysis, Investigation, Resources, Writing – review & editing, Supervision, Project administration. Kailash Shaw: Validation, Writing – review & editing, Supervision. Ketan Kotecha: Validation, Supervision, Project administration.