An Intellectual Supervised Machine Learning Algorithm for the Early Prediction of Hyperglycemia

Hyperglycemia is a chronic disease caused due to the high level of blood sugar over a long period of time. People suffering with hyperglycemia are more prone to complications such as heart disease, kidney failure, stroke and nerve damage. A machine learning based prediction model is proposed to serve the purpose of predicting this at an initial stage. Datasets employed for the prediction would be insufficient and sometimes it has outliers in it, that poses a challenge in designing an accurate model. The risk factor and severity of hyperglycemia can be reduced significantly if the precise early prediction is possible. In this paper, the dataset is considered from the UCI database repository which has eight attributes. A decision support system is proposed that uses Random Forest and Gradient Boosting classifier. Random Forest is an expansion of Decision tree, that averages and transfers the data to the nearest tree based on majority. A Modified form of Random Forest provides prediction rate of 91%. K-cross fold validation is done in modified random forest by considering 90% training set and 10% testing set. A major problem of over fitting is reduced by hyper parameter tuning, that is by estimating the number of trees and splitting of nodes. By tuning the algorithm, the accuracy obtained is 90.6%. The number of correct and incorrect predictions is studied using confusion matrix. Gradient boosting algorithm minimizes the error rate made by the previous tree. By defining N estimators for gradient boosting, the accuracy rate is increased to 89%.