Experimenting with Multimodal AutoML: Detection and Evaluation of Alzheimer's Disease

Ujjawal Shah, Saurav Keshari Aryal

Keywords: Multimodal Learning, AutoML, Alzheimer's Disease, ADReSSo Challenge, Alzheimer's Detection, Mini Mental Status Exam

Abstract: This paper describes an experiment using AutoML, AutoGluon Tabular, to discover multimodal models for MMSE regression and AD detection. Using the ADReSSo dataset, this paper reports enhanced performance in classification models and comparable performance in regression models to the baseline, achieving a significant improvement of up to 82\% accuracy on the test dataset. In contrast, the test RMSE has a marginal difference of only 0.28 compared to the baseline.