A Computational Method to Assess Post-operative Risk of Lung Cancer Patients

Abstract—Lung cancer surgery is risky such that sometime patients died after surgery. To reduce loss, we try to create a computational model to anticipate in advance the post-operative survival among the lung cancer patients using statistical and machine learning algorithms. The dataset used in our model building process is data of patients who underwent lung cancer surgery comprising of 470 records with 17 attributes. These data were collected at Wroclaw Thoracic Surgery Centre, Poland during the years 2007 to 2011. For the purpose of validating the built model, we partitioned this dataset into training set and test set with the ratio 70% : 30% and random it 10 times to obtain 10 pairs of training-test set. The training dataset is used as input to build prediction models for the post-operative survival in the lung cancer patients by applying logistic regression and support vector machine (SVM) algorithms. The obtained two models are then compared to choose the best one with the highest predictive performance based on the mean accuracy of the ten iterations. As a result of comparison using test dataset, prediction model built from the logistic regression reaches 82.38% on its average accuracy, while the SVM approach yields 75.67% of its average accuracy.