Diabetes Prediction, using Stacking Classifier

Abstract

Diabetes is a disease, which occurs due to excessive blood sugar. It has become very common nowadays. It is dependent on various factors of the human body such as Blood Sugar Level, Weight, etc. We have used one benchmark dataset, PIMA Indians Diabetes Dataset, for training and testing our model. For predicting diabetes at an early stage using the risk-based features of a person’s health, we have developed a stacking classifier, and for the same, we have stacked 6 classifiers, namely Support Vector Machine, Artificial Neural Network Classifier, Logistic Regression Classifier, Decision Tree Classifier, Random Forest Classifier and Gaussian Naive Bayes Classifier, into a single model, which as a whole, uses Logistic Regression Classification on these 6 basic hyperparameter tuned models. Also, we have compared these 6 basic models with the stacked model in terms of performance. The results obtained are satisfactory and effective in comparison to the results of already proposed methods. We have achieved accuracy of 82.68%. The results of this model will add value to additional reports, because studies on prediction of diabetes using Stacking doesn’t seem to be common, in comparison with other Machine Learning Techniques.