Wart is the skin infection medically known to be caused by Human Papillomavirus. There are several methods for treating this skin illness among which are immunotherapy and cryotherapy, the most popular methods. Before applying treatment, physicians need to identify the most effective method for every individual case. Identifying the best treatment for each case can be a daunting task; data mining can be applied to existing datasets in order to discover knowledge for easy identification of suitable treatment method for individual wart case. This study examines the use of computational intelligence in the identification of suitable treatment method for individual warts case. Specifically, the ensemble approaches in machine learning which have been found to have better prediction performance are investigated. The most common types of warts, plantar and common, were studied in the data collected from 180 patients: 90 patients managed through cryotherapy and the other 90 through immunotherapy method. Bagging, Boosting and Random Forest (RF) ensemble methods were applied to predict the response to treatment by patients. It was observed that the Random Forest approach returns the best prediction accuracy of which immunotherapy and cryotherapy methods gave 86.66% and 93.33% respectively. To investigate the generalizability of the models, the models for cryotherapy were used to predict the immunotherapy dataset while models for immunotherapy were also used to predict the cryotherapy dataset. The cryotherapy-based model of RF returned 80% accuracy and 0.48 kappa statistics while that of immunotherapy-based model of boosting returned 85.5% accuracy and 0.54 kappa statistics. Thus, the cryotherapy-based RF model is better than that of bagging and boosting while immunotherapy-based boosting model is better than others. Physicians can safely apply this model to facilitate the selection of effective treatment method for warts.
