Applying Data Mining Techniques on Thai Traditional Medicine Treatment

Abstract

Thai Traditional Medicine (TTM) has been popularly practiced in Thai culture and society since the Sukothai period. For the past decade, Thai government and private sectors have collaboratively worked to restore it to common use, after many years of inactivity. Many hospitals now combine TTM with western treatments. Data mining techniques were applied to a comprehensive dataset of TTM treatments collected between 2012 and 2015 by Bangkrathum Hospital. The original data with 56,160 records was preprocessed, cleaned and transformed to achieve a useable dataset for use with data mining techniques. Two different techniques were applied to discover symptom relationships and models. The association rule technique was applied to a 2,000 record dataset with 134 different symptoms, and the classification technique analyzed a dataset of 15,963 records. By applying the association rule technique using the a priori algorithmic approach, some symptoms that often occur simultaneously were discovered and coded in if-then type rules. Applying these rules suggests symptoms that could simultaneously occur with other symptoms. The classification techniques of Artificial Neural Network, Decision Tree and Na??ve Bayes, were applied to create models to predict likely diagnoses from symptoms recorded in new patient records.

Ten rules for symptom association prevalence were identified using the association rule approach, based on the prevalence of 8 symptoms. The construction of a Decision Tree was proven to be the best classification technique. The data mining approach applied in the study and the resulting models can be effectively used by TTM practitioners for modeling patient diagnoses based on identified symptoms.