Evaluation of Diagnosis according to Myanmar Traditional Medicine by using Expectation Maximization

Abstract

Nowadays, computer based medical system is playing a role in assisting both diagnosis and treatment. Thus, this system intends to provide information for junior traditional medicine practitioners and user who interested traditional medicine. Before evaluating, this system stores the knowledge of traditional medical experts and medical records from previous cases as training database. And, it produces the generate rules from training data set by using Naïve Bayesian Classifier. When user inputs symptoms, this system analyzes corrected diagnosis and suitable dosage. If user inputted symptoms are not evolved by NB classification, we use Expectation Maximization (EM) step that computes maximum likelihood estimation of unlabeled data. This EM step probabilistically evaluates unlabeled data by using available labeled data which is training by NB. As a result, in this paper, we evaluate corrected diagnosis and proper dosage by using semisupervised learning method (EM with NB classification) in order to improve correctness of classifier.