A Novel Method for Predicting Blood Glucose Levels in the Elderly Based on Ensemble Optimization Algorithms

Elderly diabetes patients frequently present with clinical symptoms characteristic of the geriatric syndrome. The incidence and mortality rates of complications in this demographic significantly exceed those observed in non-diabetic elderly individuals. Blood glucose levels serve as an essential diagnostic marker for diabetes. By selecting features closely associated with blood glucose levels from routine blood tests, and subsequently developing a predictive model for blood glucose, we can potentially facilitate the coordinated prevention of diabetes in elderly individuals and reduce the cost of mass-scale medical testing. This study proposes a blood glucose prediction model, PRF-SMD-PSO, based on model fusion. Blood test indicators strongly correlated to blood glucose are identified via Pearson correlation and random forest importance analysis. Three data-driven predictive models-Support Vector Machine (SVM), Multilayer Perceptron (MLP), and Deep Belief Network (DBN)are then established. These models’ predictions are fused using a Particle Swarm Optimization (PSO) algorithm. Experimental results suggest PRF-SMD-PSO outperforms individual models and shows robust prediction. Our research also indicates samples from individuals under 60 enhance prediction model robustness. Future large-scale medical data application is expected to further improve predictive performance.