Assessment of the risk factors for type II diabetes using an improved combination of particle swarm optimization and decision trees by evaluation with Fisher’s linear discriminant analysis

Abstract

Type II diabetes is one of the chronic diseases, which is the cause of death and disability in most of the countries. The objective of this research work is to apply particle swarm optimization (PSO) algorithm with improved operating parameters along with decision tree (C4.5) for accessing the risk factors for type II diabetes. The model developed with the consideration of type II diabetic risk has parameter values with normal and abnormal levels. Experimental analysis has been carried out using an improved combination of PSO with decision trees and its various splitting measures with real-world diabetic dataset. The improvement in PSO is made by proposing self-adaptive inertial weight with modified convergence logic for the particles to accelerate in the given search space. From the result analysis, it has been observed that the risk factors corresponding to diabetes are postprandial plasma glucose (PPG), glycosylated hemoglobin (A1c), mean blood glucose (MBG) and fasting plasma glucose (FPG) identified with an improved accuracy more than that of the existing methods and algorithms. The efficiency of prediction has been tested using Fisher’s linear discriminant analysis. From the inference, it has been observed that there occurs a strong relationship between the risk factors such as PPG, A1c, MBG and FPG with the risk corresponding to type II diabetes. Hence, predictive analytics using an improved combination of PSO with decision trees can also be deployed for the identification of risk factors corresponding to other chronic diseases such as coronary heart disease and kidney disease.