|
|
 |
| |
|
|
|
A Data Mining and Artificial Neural Network Approach for Autism Spectrum Disorder Detection |
|
|
|
PP: 907-921 |
|
|
doi:10.18576/amis/200405
|
|
|
|
Author(s) |
|
|
|
Hamza Abu Owida,
Mwaffaq Abu Alhaija,
Asokan Vasudevan,
Hamza A. Mashagba,
Azlan B. Abd Aziz,
Mohammad Faleh Ahmmad Hunitie,
Nor Azlina Ab. Aziz,
Suleiman Ibrahim Mohammad,
|
|
|
|
Abstract |
|
|
| In this paper, we examine how deep learning—especially facial-image–based methods—shows promise for ASD detection. However, many models are limited by centralized training, which restricts the use of diverse, globally distributed datasets and can reduce performance across demographic groups. To address these limitations, we present a privacy-preserving federated learning framework that enables medical institutions worldwide to collaboratively train ASD detection models without sharing raw patient data beyond local sites. Our framework incorporates differential privacy, secure aggregation protocols, and adaptive communication strategies to support convergence and protect confidentiality under heterogeneous data distributions. Experiments using simulated data from five medical centers demonstrate that the federated model matches centralized baselines in predictive performance while improving demographic representation and reducing bias across ethnic and cultural groups. Across diverse populations, it achieves an average accuracy of 89.2%, precision of 88.7%, recall of 89.5%, and F1-score of 89.1%, underscoring the potential for fair, confidential global diagnostic technologies for neurodevelopmental disorders. |
|
|
|
|
|
|
|
