Diabetes Mellitus risk prediction using age adaptation models

Conventional disease prediction models frequently have imbalanced or insufficient observed samples. In a real-world situation, the learned model from an imbalanced dataset may have poor generalization ability. In this paper, we use machine learning methods to investigate age adaptation in Diabetes Mellitus (DM) risk prediction. Typical machine learning algorithms for DM prediction include Linear Regression (LR), Logistic Regression, Polynomial Regression (PR), Neural Network (NN), Support Vector Machines (SVM), Random Forest (RF), and XGboost (XGB). Based on feature compensation and soft decision threshold adjustment, we propose a novel age adaptation algorithm. The results were validated using the publicly available Pima Indian Diabetes Dataset. The experimental results show that the disease risk prediction model's efficiency and accuracy are promising. The performance dropped for different age groups when using the original features. Using the compensated features, the accuracy rates for DM models were improved considerably. Age adaptation methods can be trained on one age group and adapted to another, solving the problem of data scarcity in a specific age range.

Automated summary

In this study, machine learning methods were used to investigate age adaptation in Diabetes Mellitus risk prediction. A novel age adaptation algorithm was proposed based on feature compensation and soft decision threshold adjustment. Experimental results showed that the accuracy rates of diabetes models were considerably improved using compensated features, solving the problem of data scarcity in specific age ranges.