Supervised Machine Learning-Based Comparative Study For Diabetes Prediction with Performance Tuning

Abstract

Diabetes is a condition that is gravely affecting people's health and is spreading quickly throughout the world. The necessity for scientific research on diabetes prevention, early diagnosis, treatment, and follow-up is highlighted by the growing number of diabetes patients. The diagnosis and treatment of diseases are improved by the analysis of data made possible by current technology advancements in the medical industry. Within the context of diabetes, scientists are working to provide systematic, data-based methods for diagnosing the condition. This study examines popular machine learning models for diabetes detection in order to achieve this goal. Concurrently, these models' hyper parameters were adjusted, and the impact of these adjustments on performance was examined. First, the Pima Indians data set was used to train and compare the methods for Logistic Regression, K Neighbors Classifier, Decision Tree Classifier, Random Forest Classifier, Support Vector Machine, Gradient Boosting Classifier, and LGBM Classifier. The Gradient Boosting (XGB) algorithm produced the highest accuracy values at this point for the metrics of accuracy, recall, precision, and f1-score, which were 91.03%, 91.83%, 91.77%, and 91.14%, respectively. Next, using the same dataset, all models' hyper parameters were adjusted and trained once more. The models' success increased significantly after the tuning procedure, and the Gradient Boosting approach once more produced the greatest values for the metrics of accuracy, recall, precision, and f1-score, which were 93.46%, 93.71%, 93.96%, and 93.08, respectively. © The Institution of Engineering & Technology 2024.