Enhanced Feature Selection in Lung Images by Shapley – RRO Approach

Feature selection is essential for improving the accuracy with which machine learning algorithms identify lung disorders from medical images. This research aims to integrate Shapley Values, grounded in cooperative game theory, with Raven Roosting Optimization (RRO), an advanced metaheuristic technique, to develop an enhanced feature selection methodology. Shapley Values are used to assess the global importance of features derived from X-ray and CT scan images, whereas RRO is employed to optimize the selection of the key features to train the model. Such an approach outperformed conventional techniques when tested on two benchmark datasets, ChestX-ray14 and LIDC-IDRI. The methodology performs well in the ChestX-ray14, with 91.5% accuracy, 0.90 F1-Score, a precision of about 89%, and a recall of approximately 91%. An enhancement on about the same level is reported on the LIDC-IDRI, with 93.2% accuracy, 0.91 F1-Score, 90% precision, and 92% recall.