HYBRID MODEL FOR PERFORMANCE EVALUATION AND FEATURE SELECTION APPLIED ON ENDOMETRIAL CANCER DATA

ABSTRACT Many factors affecting the success of data mining techniques, the pureness of data is one of the factors. The inclusion of irrelevant and noisy data in the pattern analyzing phase, can results poor predicting performance. To discover information from the endometrial carcinoma data set the process of cleaning, transforming and modeling are applied. Diverse kinds of pre processing techniques were functions in the data set in order to work with the full pledged data set. Methodology Used: The data mining tool Rapid Miner 5 and WEKA were used for this purpose. In this study, the endometrial cancer data was taken from the TCGA data portal with 547 patients' clinical data. Feature selection was done using various classifiers with the help of different evaluators and search method. Using R language the accuracy of the classifiers' model was checked for the minimized and full data sets. Findings: Hybrid model was adopted for the performance evaluation by combining naïve bayes and decision tree classifier and the accuracy of the new model is 92.71%. Keywords: Endometrial, R, carcinoma, classifiers, naïve bayes, decision tree.