Upgrading the Performance of Machine Learning Based Chronic Disease Prediction Systems using Stacked Generalization Technique

Abstract

During the past few years, mortality due to chronic diseases has increased manifolds globally. Low- and middle-income group countries are worst affected as the healthcare facilities are neither affordable nor accessible easily. The challenges of unaffordable and inaccessible healthcare can be faced graciously by using machine learning based prediction models. These techniques are used to learn patterns from the medical datasets and build decision support systems for diagnosis of diseases in early stages and hence prevent high mortality. This research work is aimed at upgradation of the performance of chronic disease prediction models using Stacked Generalization approach. In this work, stacked generalization ensembling approach has been applied over five base classifiers namely Logistic Regression (LR), K-Nearest Neighbours (KNN), Support Vector Machines (SVM), Naïve Bayes (NB) and Decision Tree (DT) with 10-fold cross validation. Experimental results highlight the effectiveness of Stacked Generalization method in enhancing accuracy, sensitivity, specificity, positive predictive value, and negative predictive value of prediction system by reducing variance error to avoid overfitting. The prediction models developed in this study can efficiently be used in primary health care centres to diagnose five chronic diseases namely cardiovascular diseases, diabetes, breast cancer, hepatitis, and chronic kidney disease.