Examination of ensemble learning methods in classification problems and an application on non-small cell lung cancer data

Aim: The aim of this study is to investigate and improve the usability of AI -based community learning methods in medicine. Using the data set including various clinical variables to classify NSCLC death status, NSCLC variable; The performance of ensemble learning methods in classifying NSCLC variable will be examined, and the best model will be determined. Material and Methods: This study was conducted using the data set containing various clinical variables to classify the death status of NSCLC patients on the addressed open source website. The data set includes a total of 181 inoperable stage I-IIIB NSCLC patients. Approximately 55.2 % of the data set consisted of patients receiving radiotherapy or chemotherapy, while 44.8 % consisted of patients receiving radical treatment. Individual classifiers such as SMO, K-NN, random forest and XGBoost, which are machine learning methods, and their performances, and voting, bagging, boosting and stacking methods from ensemble learning methods were used. Results: According to the general evaluation, the boosting ensemble learning method provided the highest performance in the metrics of accuracy, sensitivity, precision, specificity and the area under the ROC curve. The boosting ensemble learning method, which provides the highest classification performance with XGBoost, achieved 0.982 accuracy value, 0.971 sensitivity value, 0.989 precision value, 0.989 specificity value and 0.998 ROC curve. Conclusion: Ensemble learning classifiers gave better results in classifying NSCLC mortality according to the basic classifiers. It is recommended to use ensemble learning methods for classification problems in cancer patients with high prevalence in order to achieve successful results.