Comparative Analysis of Resampling Techniques on Predictive Model Performance for Lung Cancer Prognosis

Lung cancer has the highest mortality rates among both men and women, making it the leading cause of cancer-related deaths worldwide. Lung cancer is often detected at an advanced stage when treatment options are limited. An earlier diagnosis can increase the chances of recovery and restrict the spread of the disease throughout the body. Machine learning is increasingly utilized to predict the likelihood and presence of cancer in patients based on various factors. Predictive model performance can be affected by class imbalance in datasets, particularly when the target variable has an unequal representation of classes. This paper examines different techniques to balance data, such as Random Oversampling, SMOTE (Synthetic Minority Over-sampling Technique), Random Under sampling, and SMOTE with Tomek Links, and assesses their impact on the predictive models used for lung cancer prognosis, which include Random Forest, XGBoost, SVM, Neural Network, and Logistic Regression. It can be concluded that Random Forest exhibit superior performance in lung cancer diagnosis, achieving overall accuracy of 93.47% under Random Over Sampling technique.