Abstract:
Dementia is a significant public health issue worldwide, and early detection is important to address the coalescence of the condition. Traditional diagnostic approaches conventionally involve extensive clinical expert deliberations that may be time-consuming or subjective. This investigation investigates ensemble machine learning tactics for early dementia detection and classification. Various machine learning algorithms were used to predict the outcome of enrollments on dementia such as Random Forest, LightGBM, XGBoost and Stacking. The research indicates that Stacking ensemble models outperformed all models using an accuracy of 99.04% due to its various base learners. Random Forest- LightGBM followed with an accuracy of 98.1%, while XGBoost and Voting models had relatively lesser outputs. The vast partitions of the former models made stacking ensemble the most accurate based on predictable results, with most of the different weaknesses of the models used to generate data disposed of by others. This research has verified for the first time the potential for machine learning to supplement the expertise of pharmacologists or clinicians which enables new automation and superior diagnosis of dementia and associated characteristics. The possibilities are bound to more research to predict results in the future and interventions. The capacity of ensemble methods to overcome deficiencies in any one particular algorithm increases reliability and lowers the potential for costly misclassification. The integration of clinical decision support systems to aid providers in interpreting the model’s predictions may be a topic for future study. The hope is that the end product will be an easy to use tool for clinicians in the early identification and customized treatment of dementia
