Abstract
Worldwide, someone develops dementia every 3 seconds. Dementia is mostly brought on by Alzheimer's disease (AD). Research has concentrated on diagnosing AD and dementia over the past centuries, and brain Magnetic Resonance Imaging (MRI) has been proved an effective biomarker of AD and other dementias. Throughout the years, many methods, including various forms of Neural Networks, Support Vector Machines, and other machine learning algorithms, have been innovated and applied to the classification of brain MRI scans. This paper aims to propose a model framework that has been rarely used in this field. This novel architecture utilizes a Convolutional Neural Network (CNN) for the feature extracting task and a Random Forest (RF) model for classifying different stages of dementias. The model was evaluated on each label's performance and overall performance. The performance metrics include accuracy, f-1 score, precision, and recall. The comparisons between the proposed model and the other two sodels, a CNN and a combination of Principal Component Analysis (PCA) and RF, were also provided. The implementation of the proposed model resulted in the highest overall accuracy, weighted precision, weighted recall, and weighted f-1 score. It also guaranteed stable and excellent performance across every label.