From data to insights: Evaluating Variational Autoencoders and Generative Autoencoders for Migraine Research in AI Driver Healthcare

Abstract

This study investigates the clustering performance of Variational Autoencoders (VAE) and Generative Autoencoders (GAE) on a migraine dataset. Autoencoders (AEs) are increasingly used in unsupervised learning for their ability to reduce dimensionality and capture complex data representations. In this work, both VAE and GAE models were trained under identical conditions on a migraine dataset to evaluate their effectiveness in generating latent representations for clustering. The models' latent spaces were clustered using K-means, and performance was assessed using the Silhouette Score. Results indicate that the GAE outperformed the VAE, achieving a Silhouette Score of 0.50 compared to 0.39 for the VAE, demonstrating the GAE's superior capacity to produce meaningful latent representations. These findings underscore the importance of selecting an appropriate autoencoder architecture for healthcare applications involving clustering.