Predicting Cancer Survivability: A Comparative Study

Abstract

The prediction of cancer survivability in patients remains a challenging task due to its complexity and heterogeneity. Nevertheless, studying cancer survivability has been receiving an increasing attention essentially because of the positive impact it has on patients and physicians. It helps physicians determine the suitable treatment options, gives hope to patients, and improves their psychological state. This paper aims to predict the survival period a patient can live after being diagnosed with cancer disease by surveying the performance of three different regression algorithms. The three regression algorithms used are Decision Tree Regression, Multilayer Perceptron Regression, and Support Vector Regression. The algorithms are trained and tested on nine cancer types selected from the SEER dataset. The prediction models of each regression algorithm are built using cross validation evaluation method and ensemble method. Our experimental results show that Decision Tree Regression outperforms the others in predicting the survival period in all the nine cancer types.