Alzheimer’s Disease Classification Using the TabNet Model Enhanced by Hyperparameter Optimization

Abstract

Alzheimer’s disease is a progressive neurodegenerative disorder that leads to a gradual decline in cognitive function and remains challenging to diagnose at an early stage, as clinical symptoms often emerge after substantial brain damage has occurred. Therefore, accurate and efficient predictive models based on clinical data are essential to support early detection. Recent advances in deep learning for tabular data, particularly the TabNet model, enable adaptive feature selection through attention mechanisms while preserving interpretability. This study applies TabNet for Alzheimer’s disease classification using clinical tabular data and enhances its performance through hyperparameter optimization employing Grid Search, Random Search, and Bayesian Optimization. Model evaluation was conducted using accuracy, area under the curve (AUC), confusion matrix analysis, and execution time. Experimental results show that Random Search achieved the highest classification accuracy of 90.53%, whereas Bayesian Optimization obtained the highest AUC of 94.82%, indicating superior discriminative capability. These results demonstrate that integrating TabNet with appropriate hyperparameter optimization strategies provides a competitive, efficient, and interpretable approach for Alzheimer’s disease classification, supporting its potential application in data-driven clinical decision support systems.