Predicting Infectious Diseases Using XGBoost Algorithm and Discovering Dominant Features Using SHAP Model Interpreter

This study aimed to evaluate the performance of the XGBoost algorithm and identify the most significant features in predicting three infectious diseases based on specific geographic clusters: DHF, COVID-19, and Varicella. Understanding the characteristics of these diseases will aid healthcare workers in forecasting patient conditions using historical patient data. The methodology begins with preparing a dataset comprising 68,666 patient visits, including 2,679 patients diagnosed with the three diseases. The data source consists of patient records from a Bandung Regency clinic in the Bojongsoang District, covering three years from January 1, 2019, to December 31, 2021. The dataset was divided into 80% for training and 20% for testing. The primary data source includes keywords related to the characteristics of the three diseases in the medical history data, which were transformed into 19 features. In addition to these 19 features, other data, such as temperature, collected from medical examinations conducted by nurses during registration, were also incorporated. Experiments were performed using RapidMiner with the XGBoost algorithm, followed by an experiment using Python to evaluate the most significant features with the SHAP model interpreter. The results indicated that XGBoost achieved an accuracy of 73.32%, an average recall of 62.28%, and an average precision of 75.22%. Furthermore, the SHAP model interpretation revealed that vCough, vTemp, vSpot, vNauseous, and vFever were the most significant features for predicting the three diseases.