Enhancing Classification Accuracy for Diabetes Detection using Optimal Feature Combination

This research presents a comprehensive comparative analysis of several machine learning models applied to the task of predicting diabetes based on the well-known diabetes dataset. The dataset that is used is an open-source dataset from Kaggle having two classes, eight features and 768 samples. Among the predefined eight features (pregnancies, glucose, blood pressure, skin thickness, insulin, BMI, diabetes pedigree function, and age), a pool of different features was tested and the optimal feature combination is identified for diabetic detection based on evaluation matrices such as accuracy, precision, recall and F-1 score. The machine learning models investigated include K-Nearest Neighbors (KNN), Naive Bayes, Linear Discriminant Analysis (LDA), Decision Tree, Random Forest, Artificial Neural Network (ANN), and Support Vector Machine (SVM). The results indicate that combination containing Skin thickness and Insulin as features shows the lowest accuracies, while combinations with age, BMI and glucose show the highest accuracies. Over all the optimal feature combination of Age, BMI, glucose, blood pressure, diabetes pedigree function, and pregnancies shows the highest accuracy of 93.5% using RF with significance p<0.005. This research provides optimal feature combination with an optimal classifier for diabetic detection.