Prediction of Wart Treatment Using Deep Learning with Implicit Feature Engineering

Warts are skin growths that are caused by the human papillomavirus (HPV) through direct or indirect contact with an object used by a person with the virus. There are different types of warts and different treatment methods accordingly. Of which the cryotherapy method is preferably used to enhance the existing conventional treatment methods. As the enhanced method, the treatment decision process is very important to keep tracked and support for similar cases in future. In this study, a decision support system using machine learning is proposed to predict whether the selected wart treatment method could be successful or not using actual samples from a public dataset. There are some machine learning researches in this field using artificial neuron network (ANN) to solve the problem due to the structured dataset which is more suitable for other methods like k-nearest neighbors (kNN) or Random Forest (RF). This study uses a deep learning neuron network (DNN) approach with an implicit feature engineering method to deal with categorical features to learn implicit interactions among them. Furthermore, the k-fold cross validation is used to evaluate the proposed algorithm and the proposed model achieves the results up to 97.78% of Accuracy, 99.94% of the Area Under the Receiver Operating Characteristics Curve (AUC), 98.00% of Sensitivity and 98.00% of Specificity to predict the wart treatment method using the public cryotherapy dataset from the UCI Machine Learning Repository. This confirms that the proposed framework outperforms other methods using the same dataset.