Multimodal Data Integration for Early Alzheimer’s Detection Using Random Forest and Support Vector Machines

Abstract

Alzheimer's is a very challenging brain disease to recognize, diagnose, and treat correctly when it appears in its earliest forms. The primary contribution of this research study is about machine learning models, techniques, and approaches. In contrast, Random Forest and Support Vector Machine (SVM) are particularly suitable for identifying and staging Alzheimer's disease stages using multimodal data sources. In this paper, the aim was to develop well-performing predictive models to help diagnose Alzheimer's disease at an early stage by combining neuroimaging data (MRI/PET images), imaging-based biomarkers, both structural and functional measures from MRI(P) /PET image analysis along with subject-specific demographics like age using clinical features in a probabilistic fashion obtained from the Alzheimer's Disease Neuro-Imaging Initiative (ADNI) database. The methodology focuses on data pre-processing, feature selection, and model building using supervised learning algorithms. The accuracy of the Random Forest model is 78%, having a high performance in classifying some classes while showing different marks of performances across other courses. SVM reached an accuracy of 61%, or the model's performance is good in some classes and not reliable to identify samples from the others. The findings of this study underscore the capabilities and limits of these machine learning models in identifying Alzheimer’s disease and highlight the importance of feature engineering, data pre-processing, and model tuning to increase performance and correct class unevenness and misclassification.