Data mining models for predicting oral cancer survivability

In this paper, three predictive models are proposed to identify the most effective model for predicting the survival rate of oral cancer in patients who visit the ENT OPD. This study examined 1,024 patients who visited a tertiary care center during Jan 2004 and Dec 2009. The predictive models developed in this work are Single Tree, Decision Tree Forest and TreeBoost based on classification analysis. For all these models, it is observed that there is no misclassified row in any category and all cases have correctly been classified. The sensitivity and specificity of these models is 100 %. All the models display similar results and performance; however, as the TreeBoost model considers all 18 predictors for each split, it is marginally better than the Single Tree and Decision Tree Forest. The experimental results of probability calibration, threshold analysis and lift-gain are also slightly better in case of the TreeBoost model. Thus, the TreeBoost classification model is optimal for predicting survivability of oral cancer patients.