Implementation of Synthetic Minority Oversampling Technique and Two-phase Mutation Grey Wolf Optimization on Early Diagnosis of Diabetes using K-Nearest Neighbors

Abstract

Abstract. Diabetes is a disease attacking the endocrine system characterized by high blood sugar levels. International Diabetes Federation (IDF) estimates that there were 451 million people with diabetes globally in 2017. Without treatment, this number is expected to rise to 693 million by 2045.  One method for preventing increases in the number of diabetics is by early diagnosis. In an era where technology has developed rapidly, early diagnosis can be made with the machine learning method using classification. In this study, we propose a diabetes classification using K-Nearest Neighbors (KNN). Before classifying the data, we select the best feature subset from the dataset using Two-phase Mutation Grey Wolf Optimization (TMGWO) and balance the training data using Synthetic Minority Oversampling Technique (SMOTE). After dividing the dataset into training and testing sets using 10-fold cross validation, we reached an accuracy of 98.85% using the proposed method. Purpose: This study aims to understand how to apply TMGWO and SMOTE to classify the early stage diabetes risk prediction dataset using KNN and how it affects the results. Methods/Study design/approach: In this study, we use TMGWO to make a feature selection on the dataset, K-fold cross validation to split the dataset into training and testing sets, SMOTE to balance the training data, and KNN to perform the classification. The desired results in this study are accuracy, precision, recall, and f1-score. Result/Findings: Performing classification using KNN with only features selected by TMGWO and balancing the training data using SMOTE gives an accuracy rate of 98.85%. From the results of this research, it can be concluded that the proposed algorithm can give higher accuracy compared to previous studies. Novelty/Originality/Value: Implementing TMGWO to perform feature selection so the model can perform classification with fewer features and implementing SMOTE to balance the training data so the model can better classify the minority class. By doing classification using fewer features, the model can perform classification with a shorter computational time compared to using all features in the dataset.