Performance-optimized Alzheimer’s detection using machine learning with SMOTE and randomized hyperparameter tuning

A neurodegenerative disorder named Alzheimer's disease is a hazardous disease in the human brain that affects neurotransmitters, neurons, and tissue, hence impairing sensory perception, memory, and behavior. Currently, there is no appropriate cure approach for this disease. Nonetheless, prescribed medications can mitigate the progression of the condition. As a result, the early identification of Alzheimer's disease could play a vital role in treatment and further research. The key obstacles to the early assessment of Alzheimer's disease are the very low number of trained samples and the increasing amount of feature descriptions, while using different classification algorithms. Moreover, there are no trustworthy AI-based solutions for detecting this disease. In our study, we proposed a system in which supervised machine-learning models were implemented and deployed a web application for the initial detection of Alzheimer's disease. The Random Forest, Support Vector Machine, Decision Tree, Logistic Regression, and K-Nearest Neighbors classifiers are utilized to predict this disease. The dataset was collected from the Kaggle website and consisted of 35 features and 2149 data points. In addition, the dataset was balanced, and the feature importance was evaluated to achieve better accuracy. Among all the classification models, the RF model has achieved the highest accuracy of 91.19%.