Proceedings of the 2019 IISE Annual Conference May 18-21, 2019, Orlando, FL, USA Predicting the Performance of Cryotherapy for Wart Treatment Using Machine Learning Algorithms Md Mamunur Rahman, Shouyi Wang, Yuan Zhou, Jamie Rogers Department of Industrial, Manufacturing and Systems Engineering The University of Texas at Arlington TX 76019, USA Abstract Warts are non-cancerous tumors that can appear on the top layer of skin of different parts of the human body. For the treatment of warts, cryotherapy, a method of medical therapy that involves the application of extremely low temperatures to destroy anomalous or diseased tissue, has been commonly adopted in practice. However, the effectiveness of this treatment method varies from patient to patient. By utilizing a secondary data set which was collected from 90 patients in a dermatology clinic, this study aims to develop an accurate classification model to predict the effectiveness of cryotherapy on individual patients. To sort out the important factors, Fuzzy Entropy and Mutual Information based feature selection method has been utilized. Several machine learning algorithms have been deployed and the classification performances of these algorithms have been examined by 10-fold cross-validation method. The Support Vector Machine (SVM) with Radial Basis Function (RBF) kernel and K-Nearest Neighbors (KNN) algorithms have been found to provide promising results with an average prediction accuracy of 95.11% and 96.78%, respectively. There are several potential benefits of this study. The classification model will assist the physicians as a decision support tool to determine when to select cryotherapy over other available wart treatment methods for each unique patient. Furthermore, valuable time and hospitals’ resources can be saved by reducing readmissions and possible side effects may be avoided for some patients due to inappropriate selection of cryotherapy as a treatment process. Keywords Wart treatment, Cryotherapy, Machine Learning, Fuzzy Entropy and Mutual Information. 1. Introduction Nowadays, Machine Learning (ML) algorithms are being applied successfully in different fields of health science, e.g., disease diagnosis, selecting a treatment method, and understanding disease development. In literature, a number of studies are found related to the application of ML algorithms on various skin diseases [1–5] but the number of studies on wart treatment using ML is quite a few [6–8][9] though warts are one of the most common skin diseases for all ages of people. The virus which is responsible for warts is known as Human Papilloma Virus (HPV) [10]. Wart viruses are contagious and damaged skin can be affected by this virus by direct contact. Though sometimes warts get cured automatically without any medical intervention, it requires proper medical attention in most of the cases. There is no single treatment approach that is ideal for all patients. Some of the common medical procedures for the treatment of warts are – cryotherapy, application of salicylic and lactic acid, zinc oxide, surgical removal, immunotherapy, electrocautery, and laser ablation [11–14]. The success of these treatment methods varies from patient to patient and have various side effects. Moreover, some of these treatment methods are expensive and require several sessions for an elongated period [15]. Therefore, it is important to choose an appropriate treatment method for a patient to get effective results. In this study, we investigated the application of several ML algorithms to predict the effectiveness of cryotherapy for wart treatment on individual patients. 2. Materials The data used in this study were obtained from the UCI Machine Learning Repository [16]. The original data were collected from a dermatology clinic in Iran from January 2013 to February 2015. The records of the dataset contain information regarding the effectiveness of Cryotherapy with liquid nitrogen for the treatment of warts on 90 patients. The patients went through maximum of ten sessions with an interval of one week. In case a patient is not cured within ten sessions, the treatment procedure is changed to other methods. The dataset contains seven features which are shown in Table 1. The response variable is the outcome of Cryotherapy. Out of 90 patients, cryotherapy worked for 48 patients and for rest of the 42 patients it did not work. The other features are – the gender of the patients, age, time elapsed before starting the treatment procedure, number of warts, type of warts, and surface area of the largest wart.