An Efficient Gradient-SVM Fusion Model for Early Lung Cancer Diagnosis using Feature Engineering and SMOTE
Aruchamy Rajini, G.Paul Suthan · 2025
Cancer is a group of diseases that arise from uncontrolled proliferation of cells. Cancer cells will disrupt the natural cycle by growing excessively and uncontrollably, whereas normal cells in the human body develop, divide, and die in a regular manner. Lung cancer develops when cells in the lungs grow out of control, forming tumors and spreading to other regions of the body, including lymph nodes and surrounding tissues. Lung cancer has a significant influence on patients and their families, causing them to experience difficulties leading regular lives, including coughing, breathing issues, psychological and emotional issues. Furthermore, lung cancer has the highest death rate in the world, and the best patient outcomes and survival rates are obtained when the disease is detected early. However, considering the nature of the disease and certain limitations and inadequacies of the current diagnostic techniques, early detection of lung cancer is extremely difficult. Therefore, the model developed in this research work for the early diagnosis of lung cancer uses feature engineering, also known as feature aggregation, which combines several features from the dataset to provide more representative and informative features to improve algorithm efficiency. The dataset’s noisy and unnecessary data is eliminated using the threshold-based outlier approach. The class imbalance in the lung cancer dataset used in the present research is being addressed by the implementation of the SMOTE technique. Then, patients with and without lung cancer are categorized using the Gradient-SVM Fusion model, which combines the advantages of the Support Vector and Gradient Boosting machine learning methods. Using a variety of performance metrics, including Accuracy, Precision, Recall, the output of the proposed model is compared with that of the current models. The results show that the proposed model is the best model, with the greatest accuracy of $\mathbf{9 8. 5 \%}$.