Development Of A Model For Screening Patients With Dementia In The Early Stages with Machine Learning Techniques

Abstract

Accurate early detection of dementia is crucial for effective management. This study aims to develop and compare the performance of three machine learning models: Random Forest (RF), Logistic Regression (LR), and Support Vector Machine (SVM) for screening brain conditions. Using a dataset of 1,842 samples, including dementia data and 18 health factors, the data was analyzed, and models were developed according to the standard CRISP-DM data mining processes. Performance was assessed using accuracy, F-measure, sensitivity, and specificity. The results showed that LR had the highest accuracy (94.03%), followed by RF (93.60%) and SVM (92.56%). LR also demonstrated superior F-measure (96.76%), sensitivity (99.51%), and specificity (47.63%). This study indicates the potential of LR in dementia screening, outperforming RF and SVM in terms of accuracy, efficiency, and balanced classification. These findings support the development of machine learning-based tools to aid clinical decision-making in dementia diagnosis and effective treatment planning.