Development of Machine Learning Based Diabetes Mellitus Survival Prognostic Model

Diabetes places a heavy burden on those who have it, the healthcare system, and society at large. It is one of the major and increasingly health challenges causing serious illness to individuals with it, and which has been growing to become global crisis. This study proposed prognostic model for the survival outcome of individual with diabetes mellitus. Data with suitable attributes was collected and preprocessed using various approaches such as dealing with missing value, exploratory data analysis, 5-fold cross validation, hyperparameter tuning. The algorithms adopted includes Logistic Regression, Decision Tree, Random Forest, CatBoost, and XGBoost. Moreso, the model performance was evaluated using classification report, confusion matrix, area under curve, AUC score and receiver operating characteristics, ROC curve. Based on the results and performance comparison of the models, it was concluded that Logistic Regression with accuracy of 80% and AUC score of 84% performed better than other four algorithms used. Thus, suitable for making predictions on survival outcomes of individuals with diabetes mellitus and can be used for support system for medical practitioner.