An Improved Decision Tree Method Base on RELIEFF for Medical Diagnosis

There emerges an increasing need to mine and analyze the health data from smart home medical systems and community medical organizations. Regarding to the influence of irrelevant attributes, in this study, an improved C4.5 decision tree method based on RELIEFF attribute weighting techniques is proposed for medical diagnosis. This method includes two steps: the first step is to delete the irrelevant attributes for the classification using the RELIEFF algorithm, the second one is to build effective disaggregated model for medical diagnosis using C4.5 decision tree. Two experiments for UCI dermatology data and health examination data have been conducted respectively. Results prove that the proposed method can improve the accuracy of medical data classification for illness diagnosis, as a valuable decision support tool for doctors.