Problemy doboru zmiennych objaśniających w klasyfikacji danych medycznych
Feature selection and its impact on classifier effectiveness – case study for medical data

Summary/Abstract: The article concerns the problems of feature selection in supervised classification models for incomplete and imbalanced data. We compared the results of the application of feature selection methods implemented in the WEKA and STATISTICA machine learning environments. The impact of particular feature selection methods applied in conjunction with the pre-processing methods of missing and imbalanced date on the effectiveness and efficiency of selected single and ensemble classifiers was analyzed. The comparative analysis used updated data from the Wrocław Centre for Thoracic Surgery, on patients operated between 2006 and 2013 due to lung cancer. Sets of rules relating to hospital clinical and managerial decisions have been extracted for selected feature selection and classification methods, and for data relating to preoperative risk assessment.

• Details
• Contents