Prediction of Postoperative Survival Time of Hepatocellular Carcinoma Patients Based on Machine Learning

Abstract

 The survival rate of patients with primary liver cancer after operation is very low. Accurate prediction of postoperative survival of cancer patients plays a key role in the whole treatment period, both for patients and medical staff. For cancer patients, accurate prediction of survival time can make patients have a positive and optimistic attitude when fighting against the disease, and can also reasonably plan the rest of life to realize some wishes. For medical staff, doctors can only predict the survival time of patients through their experience before, and patients' expectations for curing diseases are generally high. Accurate survival prediction can avoid the contradiction between doctors and patients. According to the difference of prognosis of patients with clinical data, medical staff can make individualized postoperative treatment plan for patients to prolong the survival time of patients. To sum up, it is necessary to accurately predict the postoperative survival time of patients through clinical data. In this paper, survival time of 437 hepatocellular carcinoma patients was divided into 1, 3, 5 years and the unbalanced data set is oversampled and shuffled. Then 784 samples are divided into training set and test set by ten fold cross validation. Six machine learning algorithms, were used to predict the postoperative survival time of patients. The performance of the model is evaluated by calculating the accuracy. The optimal machine learning model is selected and the XGboost algorithm will be further optimized by Grid Search. Finally, the survival time of hepatocellular carcinoma patients will be well predicted.