Methodology:

• Data Collection: The study involved 228 lung cancer patients, with tumor segmentation and feature extraction performed on their pre-biopsy CT images (figure 1-3).
• Preprocessing: To address interscanner variability, ComBat analysis was used. To ensure data completeness, missing values were imputed using the mean strategy.
• Feature Selection: The initial number of features was reduced from 230 to 162 by removing features with more than 50% missing values. To select the most critical features and improve model training, the dataset was refined by Iterative Feature Elimination (RFE) and SelectFromModel methods using a Random Forest classifier. The RFE algorithm was used to incrementally train the data in the dataset from 10% to 90%. The model's performance was evaluated by accuracy and F1 score metrics at each iteration. This process resulted in selecting 30 features that performed the best and covered 20% of the results. This feature selection method improved the model's generalization capability and identified the most important features.
• Model Training: Three machine learning algorithms, Random Forest, XGBoost, and the HRFC model, were employed. The dataset was split into a training set (%80) and a testing set (%20), and models were evaluated based on accuracy and F1 score.