Collaborative Federated Learning Enhances ASD Screening Across Age Groups

Overview

This study demonstrates that personalized federated learning frameworks achieve high accuracy in screening Autism Spectrum Disorder (ASD) across children, adolescents, and adults while preserving patient privacy. Federated learning models outperform traditional centralized machine learning approaches in fairness, robustness, and efficiency under heterogeneous data conditions.

Background

Autism Spectrum Disorder (ASD) is a complex neurodevelopmental condition characterized by social communication challenges and repetitive behaviors, with an estimated prevalence of 1 in 100 children globally. Early and accurate screening is critical for timely intervention but is hindered by subjective, resource-intensive assessments and data privacy concerns. Traditional machine learning models for ASD detection often suffer from overfitting and limited generalizability due to fragmented datasets and privacy restrictions. Federated learning offers a promising privacy-preserving alternative by enabling collaborative model training without sharing raw patient data.

Data Highlights

Key Findings

• Federated learning algorithms (FedPer, pFedMe, q-FedAvg) achieved superior accuracy compared to centralized models across all age groups.
• Highest global accuracy was observed in children (97.2%), followed by adolescents (89.5%) and adults (86.8%).
• The federated framework improved fairness and robustness in non-IID (non-independent and identically distributed) data environments.
• Computational and communication costs remained efficient despite distributed training.
• Traditional classifiers like SVM, Random Forest, KNN, and J48 were benchmarked but showed lower generalizability and risk of overfitting.
• The framework supports privacy by training models collaboratively without exchanging sensitive patient data.

Clinical Implications

The proposed federated learning framework offers a scalable and privacy-preserving approach for large-scale ASD screening across diverse age groups. Clinicians and healthcare systems can leverage this technology to improve early detection accuracy while maintaining patient confidentiality and addressing data heterogeneity. This approach facilitates equitable access to ASD screening without replacing comprehensive clinical assessments.

Conclusion

Personalized federated learning represents a promising advancement for equitable, accurate, and privacy-conscious ASD screening across children, adolescents, and adults. This framework bridges machine learning innovation with ethical clinical practice to support responsible ASD detection in real-world healthcare settings.

References

1. Original Study 2024 -- Collaborative Federated Learning for Equitable Screening of Autism Spectrum Disorder in Diverse Age Groups