AUTOMATED DIAGNOSIS OF TYPE II DIABETES BY INCORPORATING MACHINE LEARNING TECHNIQUES WITH ELECTRONIC HEALTH RECORDS

Abstract

Diabetes is an irremediable disease from which millions of people are suffering all around the world Diabetes can lead to other life threating problems like kidney failure, eye sight loss etc. Due to Life style of peoples the number of individuals suffering from this disease is increasing everyday Diabetes can even lead to death. Machine learning methodologies can be helpful in detecting diabetes in infancy. In this project we have developed an ensemble model for the detection of type II diabetes. The imbalanced data is balanced using the SMOTE technique. Grid search technique is used for finding the optimal values of the hyperparameters. LightGBM and K-NN is ensemble using the soft voting classifier. The soft voting classifier will add the prediction probabilities of both the classifiers and predicts on the basis of that. The class having greater probability will be predicted by the soft voting classifier. Two datasets are used for analyzing the proposed model. Correlation graph is used for detecting the correlated features of the dataset. The proposed model gives accuracy of 90.62 for the pima Indian diabetes dataset and 94.93 for the Kaggle diabetes dataset. It is found that the proposed model performed better as compared to other state of the art models.