Predicting the risk of diabetes mellitus to subpopulations using association rule mining

Abstract

Diabetes is one of the major concerns for majority of the population. Detecting the risk of diabetes earlier in patients is essential for taking appropriate treatment at the right time. Using data from electronic medical records, risk prediction can be done with the help of association rule mining. Association rule mining generates some sets of rules which will be useful for risk prediction in subpopulations. Based on the analysis the amount of risk is estimated, and hence appropriate treatment can be done. In order to summarize the rules, four summarization techniques were analyzed and a comparison was made between them. All the four methods showed appropriate results but the last method called bottom-up summarization (BUS) was most suitable and used for accurate results. BUS technique showed the subpopulations with high risk of diabetes. A detailed analysis of the bottom-up summarization produced results with accuracy.