Applying Text Mining for Classifying Disease from Symptoms
Pannaporn Ketpupong, Krerk Piromsopa · 2018
Nowadays, misdiagnoses account for a significant portion of medical errors [1]. This is due to the fact that each physician's diagnosis is different depending on the physician's knowledge, skill, and experience. In several cases, physicians may ignore uncommon diseases. Also, after the diagnosis, the physician has to provide ICD-10-CM code. This is a difficult process for most (if not all) physicians. We propose a predictive model for classifying disease from symptoms by applying text mining technique. Our research technique allows physician to diagnose and to access an ICD-10-CM code directly from symptoms. Our models are based on several classifiers such as Decision Tree, Naïve Bayes, Support Vector Machine, and Neural Network. Models from each classifier were compared using training time, predicting time, Receiver Operating Characteristic (ROC) curve, True Positive Rate (TPR), False Positive Rate (FPR), precision and accuracy. The result suggests that Neural Network gives the best TPR at 89.03%.