Data Cleaning in Medical Procurement Database: Performance Comparison of Data Mining Classification Algorithms for Tackling Missing Value

Abstract

Data cleaning is an important process for improving the quality of decision-making information. One of today's popular cleaning tools is data mining techniques. In this paper, we focused on using data mining classification algorithms to resolve missing values in medical purchasing databases. To serve this purpose, the predictive performance of four different classifiers: Decision Tree, Naïve Bayes, K-Nearest Neighbor, and Support Vector Machine (SVM) were compared in this study. We used 2,311 medical data records from procurement database in Thailand between July 2019 and December 2019 in the experimental process. We also discussed the function of feature selection and test options that support analysis to improve model performance. The results showed that the SVM algorithm outperforms with a maximum accuracy of 89.61%. Additionally, we discussed the strengths and weaknesses of these data mining techniques for data cleaning and future research.