An optimal model for classification of lung cancer using grey wolf optimizer and deep hybrid learning

Abstract

In recent years, metaheuristic methods have shown major advantages in the field of feature selection due to its comprehensibility and possible extensive search competence. However, the majority of evolutionary computation-based feature selection algorithms in use today are wrapper approaches, which are expensive to compute, particularly for extensive biomedical data. Developing an effective evaluation strategy is crucial for significant reduction of computational cost. The proposed framework extracts deep feature from ResNet-50 and VGG-16 based convolutional neural models with initial segmentation process based on marker-controlled watershed method. Next the feature reduction is a two-fold approach with principal component analysis applied to reduce the dimensionality of large feature space from convolutional neural network (CNN) models as first step. The second step is optimal feature subset selection using a swarm intelligence method referred as modified grey wolf optimization. Finally, the selected feature subset is fed to various machine learning classifiers. The experimental result reveals that the proposed algorithm outperforms the other state-of-the-art methods with classification accuracy of 96.56%, thus upholding the dependability of the approach.