p13 / Preprint
Federated Anomaly Detection for Early-Stage Diagnosis of Autism Spectrum Disorders using Serious Game Data
arXiv:2410.20003 · 2024
In plain language
This preprint studies an autism-risk indicator from serious-game behavior using distributed anomaly detection. It explores screening support rather than establishing a clinical diagnosis.
Research context
It combines learning with limited labels and privacy mechanisms for sensitive behavioral data.
Approach
Semi-supervised autoencoders, federated learning, homomorphic encryption during aggregation, client selection, and alternative aggregators.
Findings
The experiments report that the federated approach can estimate the risk indicator while addressing data-sharing and model-aggregation privacy concerns.
Summary based on abstract. Listed as a preprint. The proposed indicator is research on ASD risk, not evidence of clinical diagnostic validation.