Early Prediction of Lupus Disease: A Study on the Variations of Decision Tree Models

Abstract Systematic Lupus Erythematosus (SLE) is an irreversible autoimmune disease that has seen to bring a lot of negative effect on the human body. It has become a very challenging task in predicting the prevalence of Lupus in patients. It has slowly gained popularity among many researchers to study the prevalence of this disease and developing prediction models that not only study the prevalence of the disease but is also able to predict suitable dosage requirements, treatment effectiveness and the severity of the disease in patients. All of these is usually done with medical records or clinical data that has different attributes related and significant to the analysis done. With the advancement in machine learning models and ensemble techniques, accurate prediction models have been developed. However, these models are not able to explain the significant contributing factors as well as correctly classify the severity of the disease. Decision Tree Classifier, Random Forest Classifier and Extreme Gradient Boosting (XGBoost) are the models that will be used in this paper to predict the early prevalence to Lupus Disease in patients using clinical records. The most significant factors affecting Systematic Lupus Erythematosus (SLE) will then be identified to aid medical practitioners to take suitable preventive measures that can manage the complications that arise from the disease. Hence, this paper aims to assess the performance of tree models by performing several experiments on the hyper parameters to develop a more accurate model that is able to classify Lupus Disease in patients in the early stages. Findings revealed that the best model was the Random Forest Classifier with parameter tuning. The most significant factor that affected the presence of Lupus Disease in patients was identified as the Ethnicity and the Renal Outcome or the kidney function of the patients.