Prediction of Diabetes Using Machine Learning and Data Mining Algorithm (Preprint)

Abstract

Introduction: Currently, diabetes is known as one of the major health problems and the most important issue in the medical profession which has a high prevalence in children and adults. On the other hand, machine learning has been introduced as a developing, reliable, and supportive technology in the field of health, and one of the interesting techniques for analyzing interventions, diseases, and conditions of the health system is the use of data mining. In fact, data mining is the process of selecting, exploring, and modeling large amounts of data.

The present study was performed in order to predict fasting blood sugar status using machine learning and data mining.

The data used in this study was from a diabetes screening program in Tehran. 3376 participants over 30 years old in 16 comprehensive health service centers participated in this screening program to check the prevalence of diabetes and its related risk factors. The dataset was not balanced according to the output variable. Therefore, the random sampling method and SMOTE technique were used for making a balance. Four different machine learning algorithms including CatBoost, Random Forest, XGBoost, and logistic regression were used to model the dataset. Also, the Shapley technique was used to select the most important features. Accuracy, sensitivity, specificity, accuracy, F1- Score, and AUC criteria were used to evaluate the model.

The results of the Shapely technique in selecting the most important features showed that the characteristics of age, waist-to-hip ratio, body mass index, and systolic blood pressure were the most important factors in predicting fasting blood sugar status. Also, the modeling results showed that the CatBoost algorithm gave the best results. For the CatBoost algorithm, various evaluation criteria including accuracy, sensitivity, specificity, and AUC were obtained as 65.98%, 71.32%, 64.54%, and 0.74% respectively.

In this study, a predictive model was developed using gradient-improved decision tree algorithms to identify the most important risk factors related to diabetes. Age, waist-to-hip ratio, body mass index, and systolic blood pressure were the most important risk factors for diabetes, respectively. This model can be used in the planning for diabetes management.