Interpretable Machine Learning for Developmental Disability Diagnosis: An Adaptive Neuro-fuzzy

This study addresses a critical challenge in pediatric healthcare: the prolonged 2- to 3-year diagnostic timeline for developmental disabilities from initial concern to confirmation. Although machine learning models show high diagnostic potential, their limited interpretability hinders their clinical adoption. To bridge this gap, we develop and validate an adaptive neuro-fuzzy inference system (ANFIS) that combines the learning capabilities of neural networks with the interpretability of fuzzy logic systems. Our longitudinal observational study analyzed comprehensive developmental assessment data from 500 children aged 2-5 years across 12 centers. The dataset encompassed cognitive, behavioral, motor, and social interaction scores, alongside demographic and family history variables. The ANFIS model was implemented using Gaussian membership functions and Sugeno fuzzy inference, incorporating age adjustments and cross-domain interactions (specifically, ratios between cognitive and social and motor-behavioral developmental domains). The model achieved 92.0% accuracy, 87.5% sensitivity, 97.6% specificity, and an area under the curve of 0.925 on a test set of 100 cases with rapid convergence by epoch 10. Feature importance analysis identified Cognitive_Social_Ratio as the most influential diagnostic indicator, underscoring the significance of cross-domain developmental relationships in clinical assessments. The ANFIS approach successfully balances diagnostic accuracy with clinical interpretability, thereby addressing a fundamental barrier to artificial intelligence (AI) adoption in healthcare. The transparent decision-making process, as evidenced by interpretable membership functions and clear feature importance rankings, enables clinicians to understand and trust the system’s recommendations. These findings suggest that interpretable AI systems can enhance diagnostic consistency, reduce assessment delays, and facilitate early access to interventions for children with developmental disabilities.