Abstract

In the recent modern world, most of the human beings are affected by diabetes. Diabetes mellitus is a silent killer which may destroy the organs in the body without being noticed. Now many researchers are developing model to predict diabetes at early stage to prevent from other health complications caused by diabetes. Most of the existing works focused on weighted KNN rather than normal KNN to obtain better performance. In this work, instead of using normal KNN or weighted KNN for doing classification, weighted dataset is computed, significant features are identified and KNN is used for predicting type 2 diabetes patients. The weight is calculated for each feature value by calculating probability from its neighbourhood. N2PS pruning algorithm is used for identifying significant features. The oversampling technique SMOTE is applied to balance the dataset when it is imbalanced. Prediction accuracy of the intended method is found as better when weighted dataset is used for prediction.

Authors

A. Suriya Priyanka, T. Kathirvalavakumar
Virudhunagar Hindu Nadar’s Senthikumara Nadar College, India

Keywords

Prediction, Network Pruning, KNN, Weighted Data, IQR, SMOTE