Important variable selection techniques with multiple solutions for medical information applications

Abstract

The purpose of this study is to utilize the strategy of multiple solutions with easily operated tools to make compatible and affordable trustworthy results. This strategy can satisfy general users in extracting specific knowledge (e.g. diagnosis, treatment, health education, hospital administration, etc.). Wide application is one of the key promoters of a successful medical information system. Risk factors of heart disease can be identified by important variable selection techniques. Four techniques, Chi-square test, correlation analysis, stepwise discriminant analysis, and decision trees, were used in this study. One must compile the results of the different methods to deal with specific research questions in a logical way. It is expected that this approach may reduce the uncertainties obtained from a single method. When applying the important index derived from the four different method results to several applications, all results have been confirmed by current medical knowledge or otherwise provide meaningful information. The prime purpose of this study is not disease diagnosis, but to enable the general health practitioner (medical doctor, hospital administrator, etc.) to formulate their hypothesis.

• Data Mining
• Multiple Solutions
• Chi-square Test
• Discriminant Analysis
• Correlation Analysis
• Decision Trees