A Hybrid mRMR-RSA Feature Selection Approach for Lung Cancer Diagnosis Using Gene Expression Data

Abstract

Worldwide Lung cancer is the leading causes of cancer-related death, thus emphasizing the need for early and accurate detection to improve patient outcomes. While imaging modalities such as Computerized Tomography (CT) are widely used for identifying abnormal tissues and tumor characteristics, integrating advanced computational methods offers transformative potential in diagnostics. This study focuses on leveraging a hybrid machine learning approach for lung cancer classification using microarray gene expression profiles. Gene expression profiling provides critical insights into genetic abnormalities associated with cancer, but the high dimensionality of the data relative to the sample size poses significant analytical challenges. To address this, a hybrid Minimum Redundancy Maximum Relevance (mRMR) and Recursive Feature Selection Algorithm (RSA) framework was developed to enhance feature selection and classification accuracy. The K-Nearest Neighbor (KNN) algorithm demonstrated superior performance, achieving high accuracy and notable improvements in precision and recall metrics. Among various models evaluated like SVM, ANN, the K-Nearest Neighbor (KNN) algorithm determined to give superior performance with achieved high accuracy of 92.37% with dataset1 and 92.01% with dataset2. These findings highlight the promise of hybrid machine learning techniques in early prediction for diagnosis, paving the way for more personalized and effective lung cancer detection and treatment strategies. The potential implications of the findings for personalized lung cancer detection and treatment are significant and transformative. The use of hybrid machine learning techniques enables earlier detection of lung cancer. This could lead to improving survival rates, Personalized Treatment Plans, Precision Medicine, Predictive Capabilities, Cost-Effectiveness.