Interpretable Machine Learning Models for Alzheimer's Disease Prediction

Alzheimer's disease is a progressive neurological disorder that damages brain cells, leading to irreversible memory loss, commonly known as dementia. It affects primarily elderly people, particularly those 65 years and older. Since the disease is incurable, it causes numerous fatalities each year. However, early detection can help slow its progression and improve patient management. An automated system is essential for early diagnosis, as it can accurately detect and classify Alzheimer's disease into different stages, enabling timely intervention. This study focused on predicting the onset of Alzheimer's disease. The features were normalized, and feature selection was performed using a correlation map and Principal component analysis (PCA). The selected feature set was then used to train various fine-tuned machine learning (ML) models. Among them, the Random forest (RF) classifier achieved the highest accuracy of 97.4 % on the train set and 93.7% on test cases. Explainable artificial intelligence (XAI) techniques, including Shapley Additive Explanations (SHAP) and Permutation Importance (PI), were utilized to interpret the decision-making process of ML models. These methods analyzed the role of different characteristics in predicting the onset of Alzheimer's disease. In particular, both explainability approaches consistently identified the same five main influential characteristics, reinforcing their significance in early diagnosis.