The Proposed Clustering Model to Predict Autism Spectrum Disorder in Toddlers

Area of Science:

• Machine Learning
• Developmental Neuroscience
• Computational Psychiatry

Background:

• Autism spectrum disorder (ASD) is a complex neurological condition requiring early diagnosis for effective intervention.
• Traditional ASD diagnostic methods can be time-consuming and expensive.
• Machine learning offers potential for faster, more cost-effective ASD screening.

Purpose of the Study:

• To develop and evaluate a robust clustering model for the early prediction of ASD in toddlers.
• To compare the performance of parametric and non-parametric clustering algorithms for ASD prediction.

Main Methods:

• Utilized five parametric and five non-parametric clustering algorithms on a Kaggle dataset.
• Evaluated algorithm performance using metrics like Silhouette Score, Davies-Bouldin Index, and adjusted rand index.
• Assessed clustering performance and processing time for each algorithm.

Main Results:

• The Gaussian Mixture Model (GMM) demonstrated superior performance over other algorithms.
• GMM achieved high clustering performance metrics (e.g., ARI = 0.9730) and time efficiency (4.1998s avg).
• When data was split, GMM achieved perfect scores (ARI=1, FMI=1, NMI=1).

Conclusions:

• GMM shows significant potential as a tool for early ASD prediction in toddlers.
• Machine learning models like GMM can provide faster and more cost-effective alternatives to traditional ASD diagnostics.
• Further validation is necessary before GMM can be considered a diagnostic tool.