Prediction and validation of survival rate of metachronous second primary lung cancer patients using machine learning classifiers

ABSTRACT Machine learning (ML) has been applied recently to develop prognostic classification models that can be used in individual cancer patients to forecast outcomes. Here, four different ML algorithms were built to predict survival rate of lung cancer patients using 1600 metadata records. Of note, the generated models were validated using test set and external validation data set consisting of 400 patient records each together with 10-fold cross-validation technique. The extratree classifier algorithm was employed to identify the influential descriptors for patients survival after incidence of metachronous second primary lung cancer. The models were assessed using five different performance metrices. The results from our study highlight that logistic regression model with all features and important features achieved an accuracy of 94% and 96%, respectively, for stratifying the survival status of lung cancer patients. On the other hand, logistic regression also outperformed external validation with an accuracy of 85%. Indeed, the results from our study will provide meaningful insights for the treatment and management of large community of lung cancer patients.