نموذج التنبؤ والتصنيف لمرض السكري باستخدام التعلم الآلي

Prediction and Classification Model of Diabetes Using Machine Learning

Abstract

Diabetes is one of the most prevalent chronic diseases globally, affecting millions and leading to serious complications such as heart disease, kidney failure, and vision loss. Early detection and proper classification of diabetes types (Type 1 and Type 2) are essential for effective treatment planning and long-term disease management. Early detection and prediction of diabetes can greatly improve patient outcomes, making it a global health concern. Manual patient data analysis is a common method of traditional diagnostic techniques, but it can be laborious and prone to human error. The research problem addressed here is the inefficiency and inaccuracy of traditional diagnostic methods, which this study aims to overcome using automated, data-driven approaches. This study investigates the use of clinical and demographic data in conjunction with machine learning (ML) techniques to predict and categorize diabetes. The study uses a variety of datasets features, such as lifestyle factors, biometric data, and clinical records, to train and assess different machine learning models, like decision trees, support vector machine (SVM), K-nearest neighbors (KNN), logistic regression, random forest, bagging, voting, Naïve bayse, XGBoost(Extreme Gradient Boosting), LightGBM (Light Gradient Boosting Machine), to create predictive models.
Three datasets used in this study include two well-known datasets sourced from publicly available repositories like the Diabetes Health Indicators containing 253,680 samples, the LMCH Diabetes dataset containing 1,000 patient samples, and a real-world dataset collected from the Imam Hassan Al-Mujtaba Diabetes and Endocrinology Center in Iraq, Karbala Governorate, containing 1,596 patient samples, includes features such as age, BMI, glucose levels, blood pressure, and insulin levels. Preprocessing techniques, including handling categorical variables, handling missing values, removing duplicates, feature scaling, addressing class imbalance and feature selection are applied to enhance model performance. The models are evaluated using confusion matrix, accuracy, precision, recall, F1-score, and AUC. According to the results, Bagging achieved an accuracy of 89.27%, while KNN achieved an accuracy of 86.9%, which are the highest on Dataset 1. For Dataset 2, the decision tree achieved an accuracy of 99.50%. Using a real-world dataset, Random Forest and LightGBM achieved the highest accuracy of 99.79% in predicting diabetes. Additionally, the voting classifier achieved an accuracy of 95.74% in identifying the type of diabetes, confirming its effectiveness in classification tasks.
This study’s innovation lies in combining real-world and large-scale datasets with advanced ensemble learning methods for both prediction and type classification. The results demonstrate that machine learning offers a scalable, efficient, and highly accurate solution for early diabetes diagnosis and classification, contributing valuable insights to the field of intelligent healthcare.