Feature Selection Based on Multiple Correlation Measures for Medical Examination Dataset

The analysis of medical data is not an easy task for health care systems since a comprehensive medical examination is performed with hundreds of parameters. These data come with considerable amount of irrelevant and redundant features. But only a subset of these features are useful for prediction, feature selection plays a significant part in to identify critical features for building prediction models. However, a number of existing filter feature selection methods can cope with redundancy and irrelevance by using single correlation measure, and ignore the interaction between features as well. In this paper, the idea of using multiple correlation measures is adopted for filter feature selection, additionally, the used method is also take feature interaction into account. Two correlation measures that converted to the standard form of 0-1 are fused together in the used algorithm, and the complementary correlation between candidate feature and selected feature can be distinguished by introducing an item. According to four common classifiers and a medical examination dataset, the experiments can be developed in order to demonstrate the validity of the method. Experimental results can confirm that the method is superior to four representative filter feature selection methods