Prediction Method of Type 2 Diabetes Mellitus Based on a Combination of Hybrid Feature Selection and Random Forest

Type 2 diabetes mellitus(T2DM) has become a major social problem threatening the health of the population; the ability to predict its prevalence can help in prevention and early treatment. Existing prediction methods face difficult discovery of predictive T2DM risk factor features and low accuracy. To address these shortcomings, we propose a T2DM prediction method based on a combination of hybrid feature selection and random forest. First, an algebraic combination of original features is used to construct a candidate feature set of risk factors. Second, the RReliefF method is used to screen the maximum relevant features to obtain the maximum relevant feature set, and the mRMR algorithm is used to eliminate redundant features to obtain the maximum relevant minimum redundant feature set (important feature set). Again, the key feature set is obtained by causal replacement of the important feature set. Finally, a diabetes prediction model is constructed using the key feature set and the random forest algorithm. Experimental results show that the model helps to screen the features with predictive contribution and effectively improves the accuracy of diabetes prediction, which can provide some reference for T2DM prevention and treatment research.