Diabetes Prediction Using Machine Learning
Chreesk Sabah M. Ali *
Department of Information Technology, Akre University for Applied Sciences, Technical College of Informatics, Duhok, Kurdistan Region, Iraq.
Omar Sedqi Kareem
Department of Public Health, College of Health and Medical Technology, Shekhan, Duhok Polytechnic University, Duhok, Kurdistan Region, Iraq.
*Author to whom correspondence should be addressed.
Abstract
Diabetes mellitus is a persistent metabolic condition impacting millions globally. Preventing problems requires early detection. This study uses a clinical cohort from Medical Centre Chittagong, Bangladesh, and the Pima Indian Diabetes dataset to create machine learning-based classification models for diabetes prediction. Five supervised algorithms, including k-nearest neighbours, naïve Bayes, support vector machines, decision trees, and multilayer perceptron’s, were trained and validated using ten-fold cross-validation following thorough data pre-processing and the selection of nine essential features. Performance measurements encompassed accuracy, precision, recall, F-measure, and area under the ROC curve. Model accuracies varied between 81.1% and 97.6%, whereas ensemble techniques had a dependability of up to 98.7% and an AUC of 0.95. These results show how integrated machine learning pipelines can help clinicians make clinical decisions when it comes to diabetes risk screening.
Keywords: Diabetes mellitus, early detection, machine learning, classification models, ensemble techniques, Pima Indian dataset, cross-validation, feature selection, ROC AUC, clinical decision support