Feature selection and comparision of classification algorithims for survival of breast cancer patients

Breast cancer has become one of the most common diseases, especially among women, increasing the importance of predicting survival. In this study, the success of machine learning algorithms on survival prediction was compared using the Surveillance, Epidemiology, and End Results (SEER) breast cancer data set and the effect of attribute selection on the success of these algorithms was examined. In the first stage of the study, 4 different machine learning algorithms were run on this data set in Waikato Environment for Knowledge Analysis (Weka) using all of the attributes that might be meaningful for survival estimation in the data set. These are: Naive Bayes, J48, Multiobjective Evolutionary Fuzzy Classifier (MEFC), Support Vector Machines (SVM). The most successful algorithm is J48 algorithm. In the second stage of the study, the classification algorithms used in the first stage were tested on the data sets obtained by using filter and wrapping attribute selection methods together and their successes were compared. It has been observed that higher achievements can be achieved with less attribute. Keywords: Breast cancer, Feature selection, Survival, Classifications algorithms