Optimized Machine Learning Models for Early Alzheimer's Disease Detection: A Feature-Reduction Approach

Alzheimer's Disease (AD) is a neurodegenerative disease that affects an individual's memory, cognitive abilities, and daily function-skills, progressive in nature. Early diagnosis of AD has implications for early intervention and management. In this context, the Alzheimer's Disease Dataset was utilized in the current study, comprising 2,149 instances and 35 features to construct predictive models using different Machine Learning (ML) techniques. In order to increase the relevance of features for better performance of the models, the dataset was preprocessed rigorously and retained only the top 12 most correlated attributes. The different machine learning models like Decision Tree, Random Forest, CatBoost, and a Voting Classifier were trained and evaluated by various metrics such as accuracy, precision, recall, and F1-score. Among the tested models, CatBoost provided the highest accuracy of 96%, outperforming other algorithms. This demonstrates the potential of feature reduction in maintaining the performance of models. The result emphasizes the ability of ML to enhance early detection of Alzheimer's disease and provides the ground for a future improvement of AD diagnostic methodologies.