| Sumario: | Lung cancer (LC) remains one of the most challenging malignancies to diagnose accurately due to the complexity of its pathology and the limitations of traditional diagnostic protocols. This study presents a novel deep learning–based ensemble technique aimed at enhancing the automatic detection of LC, integrating machine learning (ML) methodologies to improve diagnostic accuracy and reliability significantly. The publicly available Kaggle LC survey dataset, comprising 309 instances with 16 clinical attributes, including age, smoking history, and symptoms were used. The synthetic minority oversampling technique (SMOTE) for data balancing, complemented by fivefold cross‐validation strategy to assess the performance of various ML models, was also employed. This approach utilizes convolutional neural networks (CNNs) integrated with ensemble learning to develop two novel models, ConvXGB and ConvCatBoost, which synergistically combine CNN‐based feature extraction with the powerful classification capabilities of XGBoost and CatBoost, respectively, for binary classification in LC detection. ConvXGB achieved improved results, including an accuracy of 98.26%, precision of 98.72%, recall of 98.72%, and F1‐score of 98.72%. Such performance metrics underscore ConvXGB's potential to minimize false positives and negatives, enhancing outcomes in LC diagnostics. Additionally, the complexities surrounding LC diagnosis through a comprehensive comparative analysis were explored, establishing benchmarks for evaluating diagnostic precision, accuracy, recall, and F1‐score. Our findings highlight the transformative impact of ML in LC research, paving the way for diagnostic frameworks that promise significant advancements in oncological practices.
|
|---|