机器学习预测癌症五年存活率

Machine Learning Predicts Five-Year Cancer Survival Rates

癌症生存率对于癌症患者的临床治疗具有重要的意义,本篇论文旨在探究出可以准确预测癌症患者五年生存率的机器学习方法。采用的数据特征是TCGA网站上下载的多组学数据。我们探究出mRMR特征选择法和逻辑回归分类器以及SVM分类器的方法组合可以使五年存活率的准确率达到0.85以上,甚至可以超过0.9。由于我们分类时采用的是五折交叉验证,可以表明我们的结果稳健性较高。同时这两种方法组合的结果中AUC值和F1值也比较高,再次证实了这两种方法组合的优势。

Cancer survival is of great importance to the clinical management of cancer patients and the aim of this thesis is to explore machine learning methods that can accurately predict the five-year survival rate of cancer patients. The data features used are multi-omics data downloaded from the TCGA website. We explore that the combination of the mRMR feature selection method and the logistic regression classifier and SVM classifier can result in an accuracy of more than 0.85 and even more than 0.9 for the five-year survival rate. Since we use a five-fold cross-validation for our classification, our results are robust. Also the AUC and F1 values are higher in the results of the combination of these two methods, which again confirms the advantages of the combination of these two methods.