Progressive Early Alzheimer’s Diagnosis: Leveraging Hard and Soft Ensemble Voting Techniques in Machine Learning

Alzheimer’s Disease (AD) is actually the most common form of dementia characterized by a slow decline in cognitive functions, leading to damage in memories, behavioral changes, and loss in communication functions. This disease is characterized by slow progression in neurological functions, which leads to an increase in brain cell death, thereby ensuring the need for early diagnosis. This research is based on a sample consisting of 150 individuals ages ranging from 60 to 96, with the variables of age, sex, “Clinical Dementia Rating” (CDR), “estimated Total Intracranial Volume” (eTIV), “Normalized Whole Brain Volume” (nWBV), “Atlas Scaling Factor” (ASF), “Years of Education” (EDUC), “Socioeconomic Status” (SES) and “Mini-Mental State Examination” (MMSE) scores. To overcome the challenge of classifying low sample sizes and noise in the MRI images, we propose a voting ensemble approach that utilizes the predictions from a greater number of classifiers to improve accuracy and reliability. We applied the individual machine learning (ML) models and boosting techniques including “Decision Tree”, “AdaBoost”, “Gradient Boosting”, “Gaussian Naive Bayes”, “Support Vector Machine”, “XGBoost”, and “Random Forest” on the same dataset to compare their performance with the voting ensemble technique. The accuracies of individual ML models range from 90.67% to 96%. The ensemble model (“hard” and “soft voting”) achieved the accuracies of 94.67% and 96.57%, respectively. This research work demonstrates that ensemble learning greatly improves the early detection and classification of AD.