Medical knowledge based data mining for cardiac stress test diagnostics

Though an important preliminary process to diagnose cardiac abnormality, cardiac stress test is often inaccurate and involves costly follow-up procedures. Data mining based approaches have potentials to reduce the uncertainties in this respect. However, the exploration on such data mining approaches is still limited. Further, traditionally employed data mining techniques like feature selection often ignore features which are medically important, and as a consequence the results are doubted by the medical practitioners. The objective of this research is to compare the effectiveness of Medical Feature Selection against the traditional Computer-Automated Feature Selection mechanism employed by data mining community for a real life stress test scenario. Considering six different classification algorithms, it's noted that the medical knowledge motivated feature selection (MFS) improved performance of the classifiers in terms of majority of the measures, as compared to the computer-automated feature selection (CFS). While a unique classifier was not observed as the best technique, MFS had the most notable effect on Support Vector Machine (SMO) and Decision Tree (J48) for the dataset. The findings are expected to assist better computer-aided heart disease diagnostics, and is thus of interest to medical practitioners and researchers.