Safe Prediction of Diabetes Mellitus Using Weighted Conglomeration of Mining Schemes

Diabetes mellitus (DM) is a chronic disease, from which most people are suffering in the present day. In this work, an attempt has been made to propose an ensemble of numerous techniques for the diagnosis and analysis of diabetic Mellitus data. The prime focus has been given to a safe prediction. One of the reasons for focusing on this ailment is due to its increasing rate of co-existence and occurrence over the world. This ailment is causing approximately one and a half million casualties every year. This work is meant to mitigate the challenge of early prediction of DM. For this, four different data mining techniques have been utilized namely lazy K-star, multi-layer perceptron (MLP), logistic regression and random forest. Using these four techniques a conglomerate algorithm was proposed which gives the final predicted label. If a patient test data is assigned a positive DM label by more than three classifiers then only it is assigned with the final label as positive DM, otherwise, it is treated as a negative. Satisfactory results in terms of the overall rate of accuracy have been obtained through this ensemble approach. Here, accuracy refers to the correct classification and prediction of DM through the proposed scheme. The proposed algorithm obtained an overall accuracy of 98.25%.