Background: Adherence to treatment is required to ensure the best glycemic control and avoidance of complications associated with diabetes among patients with type 2 diabetes. This study aims to predict treatment adherence among Iraqi patients with type 2 diabetes using interpretable machine learning algorithms to identify the key factors influencing adherence.
Methods: This was a cross-sectional study done in 2026 to come up with predictive models of treatment adherence among patients with diabetes in Mosul, Iraq. The study sample used was 850 subjects and comprised demographic, clinical, and Trans-Theoretical Model construct variables. Machine learning algorithms and data analysis were done using the Random Forest, Support Vector Machine (SVM), and Multi-Layer Perceptron (MLP) after preprocessing the data. The dataset was further stratified into training and testing groups and the performance of the model was measured in terms of Accuracy, Precision, Recall, F1-score, and the Area Under the Receiver Operating Characteristic Curve (AUC). Moreover, the SHAP approach was used to determine the most significant predictors of treatment adherence to enhance the interpretability of the model.
Results: The best predictive model out of the developed models was the Random Forest algorithm, with an accuracy of 60.6, a sensitivity (recall) of 82.2, and an F1-score of 71.2 on the test data. Cross-validation of the Random Forest model gave a hyper parameter optimization of 0.676. The interpretability analysis based on the SHAP algorithm revealed that the most significant predictors of adherence to treatment were HbA1c, body weight, age, income, and self-efficacy and perceived barrier constructs.
Conclusion: These results suggest that machine learning algorithms, especially decision tree-based models like Random Forest, with the use of model interpretability strategies, have significant potential in predicting treatment adherence in diabetic patients.