An Experimental Comparison of Different Machine Learning Algorithms for Detecting Type 2 Diabetes

Abstract

Artificial intelligence (AI) is transforming healthcare, with (ML) offering significant potential for early disease diagnosis and personalized treatment, particularly in diabetes management. This capability is crucial for enhancing diagnostic accuracy. However, accurately predicting type 2 diabetes remains challenging due to diverse patient populations and complex data, as traditional methods can be slow and miss subtle early indicators, leading to delayed interventions. There is a clear need for robust predictive models. This paper proposes and evaluates various ML algorithms and data mining techniques for effective type 2 diabetes prediction. We utilized a unique Iraqi dataset with comprehensive features including HbA1c, lipid profiles, age, and BMI. Our methodology involved rigorous data preprocessing, including cleaning and handling class imbalance with SMOTE. Eight ML algorithms were compared: Decision Tree (DT), Naïve Bayes (NB), Random Forest (RF), Gradient Boosting (GB), K-Nearest Neighbor (KNN), Logistic Regression (LR), and Artificial Neural Networks (ANN). Model performance was assessed using accuracy, precision, recall, F1-score, MAE, RMSE, and AVE, with cross-validation ensuring robustness. The DT algorithm achieved the highest performance, with an accuracy of 99.44%, outperforming all other models. This highlights DT’s effectiveness and its potential for accurate early diagnosis, confirming HbA1c, age, and BMI as key predictors. Future work should establish a national health database in Iraq and explore advanced deep learning techniques, such as Convolutional Neural Network and Recurrent Neural Network.