Comparison of Machine Learning Models for Diabetes Prediction

Abstract

The prevalence of chronic diabetic disease has significantly increased recently. Blood sugar levels rise with diabetes, which also causes additional issues like blurred vision, kidney failure, nerve damage, and stroke. Early diabetes detection helps guide the implementation of the necessary measures. Everyone's attention is being drawn to the sharp rise in the number of diabetics. Different models have been built in this study to categorize diabetic and non-diabetic individuals. The classification models for the PIMA Indian Diabetes dataset have been implemented using machine learning algorithms likeLogistic Regression (LR), K-Nearest Neighbors (KNN), Random Forest(RF), and Support Vector Machine (SVM). Deep learning perspective algorithm such as Multi Layered Feed Forward Neural Network (MLFNN) also been implemented and comparisons were made. For better comparisons, accuracy and execution times for each algorithm are recorded. To further improve the diabetes dataset's classification accuracy, various activation functions, learning algorithms, and approaches to deal with missing information are taken into account. The results of MLFNN are then contrasted with machine learning algorithms. MLFNN has the highest achieved classification accuracy (92%) of all the classifiers and it will be more accurate if it is implemented in larger datasets. These models are built to improve the standard of the patient care. This research is helpful in predicting pre-diabetes and identifying the risk factors linked to the development of diabetes from clinical data.