[Paper Review] Accurate Autism Spectrum Disorder prediction using Support Vector Classifier based on Federated Learning (SVCFL)

Ali Mohammadifar, Hasan Samadbin+1 more|arXiv (Cornell University)|Nov 8, 2023

Autism Spectrum Disorder Research4 citations

TL;DR

This paper proposes SVCFL, a federated learning framework combining Support Vector Classifier (SVC) with decentralized data training to predict Autism Spectrum Disorder (ASD) with 99% accuracy. By aggregating four diverse datasets without centralizing raw data, the method achieves a 13% improvement over baseline models, enabling privacy-preserving, early ASD diagnosis using AI.

ABSTRACT

The path to an autism diagnosis can be long and difficult, and delays can have serious consequences. Artificial intelligence can completely change the way autism is diagnosed, especially when it comes to situations where it is difficult to see the first signs of the disease. AI-based diagnostic tools may help confirm a diagnosis or highlight the need for further testing by analyzing large volumes of data and uncovering patterns that may not be immediately apparent to human evaluators. After a successful and timely diagnosis, autism can be treated through artificial intelligence using various methods. In this article, by using four datasets and gathering them with the federated learning method and diagnosing them with the support vector classifier method, the early diagnosis of this disorder has been discussed. In this method, we have achieved 99% accuracy for predicting autism spectrum disorder and we have achieved 13% improvement in the results.

Motivation & Objective

• To address the challenge of delayed and inconsistent autism spectrum disorder (ASD) diagnosis by leveraging AI for early detection.
• To overcome data privacy and sharing barriers in healthcare by using federated learning to train models without centralizing sensitive patient data.
• To improve diagnostic accuracy for ASD using a Support Vector Classifier (SVC) trained across multiple decentralized datasets.
• To demonstrate that federated training with SVC can outperform centralized baselines in ASD prediction while preserving data confidentiality.
• To provide a scalable, privacy-preserving AI solution for early ASD screening in clinical settings.

Proposed method

• The proposed SVCFL framework uses federated learning to train a Support Vector Classifier (SVC) across four independent autism datasets without sharing raw patient data.
• Each client (institution) trains a local SVC model on its own data, then transmits only model updates (weights and gradients) to a central server.
• The central server aggregates these updates using FedAvg (FedAvg aggregation) to form a global model, which is then redistributed to clients for further training.
• The method ensures data privacy by never transferring raw data, relying only on model parameters across the federation.
• The final global model is evaluated on a centralized test set to measure performance, with hyperparameters tuned via cross-validation.
• The approach leverages the robustness of SVC in high-dimensional spaces and the scalability of federated learning for medical AI applications.

Experimental results

Research questions

• RQ1 Can a federated learning-based SVC model achieve high accuracy in predicting Autism Spectrum Disorder without centralizing sensitive patient data?
• RQ2 How does the performance of the proposed SVCFL framework compare to centralized training baselines on ASD prediction tasks?
• RQ3 To what extent does federated learning improve model generalization and privacy in multi-institutional ASD diagnosis?
• RQ4 What is the impact of combining SVC with federated learning on diagnostic accuracy and model convergence in ASD screening?
• RQ5 Can the proposed method detect early ASD signs with high sensitivity and specificity across diverse datasets?

Key findings

• The SVCFL framework achieved a diagnostic accuracy of 99% in predicting Autism Spectrum Disorder across four federated datasets.
• The method demonstrated a 13% improvement in prediction accuracy compared to baseline models trained on centralized data from the same datasets.
• The federated training process preserved data privacy by ensuring no raw patient data was shared across institutions.
• The model maintained high performance across diverse data distributions, indicating strong generalization across different clinical settings.
• The use of SVC in a federated setting proved effective in high-dimensional, small-to-medium-sized medical datasets typical in autism research.
• The results suggest that SVCFL is a viable, privacy-preserving solution for scalable and accurate early ASD screening in real-world healthcare environments.