Semi-automatic Classification Based on ICD Code for Thai Text-Based Chief Complaint by Machine Learning Techniques

Abstract

We proposed the methods to classify the text-based chief complaint in Thai language, our native language, into the symptom code based on ICD-10. Using Thai sign and symptom descriptions from ICD-10 document is the training data to build Thai text-based corpus in domain of sign and symptom. Then the corpus has been used for tokenization of Thai text-based chief complaint (ThCC) into a particular word by using the longest matching technique and our proposed technique named two-level tokenization technique. The tokens from two techniques are evaluated by five different classifiers including decision tree classifier, K-mean neighbours classifier, radius neighbours classifier, random forest classifier, and extremely randomized tree classifier. The experimental result shows 85% accuracy for assigning ICD-10 code to Thai text-based chief complaint by using our proposed technique with decision tree classifier.