Publication

Abstract : The primary focus of this work is the utilization of machine learning classifiers to forecast the likelihood of survival in individuals with lung cancer who have undergone significant lung resections via thoracic surgery. Since the data is highly imbalanced, the variations in the synthetic minority oversampling technique (SMOTE), Random Oversampling, and near-miss under-sampling were used to prevent biased decisions in favor of the majority class by the machine learning classifier. Five-Fold Cross-validation is performed before and after balancing data for all the classifiers. The Random Oversampling method outperforms the majority of machine learning classifiers. Support Vector Classifier provides the highest evaluation metrics for the following data, with an accuracy of 85% for both imbalanced and balanced (B-SMOTE-SVM) data. Since the data is imbalanced, we considered the F1 score as a measure for evaluating all the classifiers while applying various data balance techniques, with the lowest score being zero and the highest being 0.34. By combining the benefits of classifiers along with Over sampling and Under-sampling and the identification and interpretation of risk factors, this proposed solution can assist medical practitioners in their diagnosis of the survival status of lung cancer patients.