Dual feature selection and rebalancing strategy using metaheuristic optimization algorithms in X-ray image datasets

The imbalance and multi-dimension are two common problems in the medical image datasets, which affect the performances of the image processing procedures. The traditional methods to solve these two problems are notoriously difficult. Accordingly, this work employed metaheuristic methods to optimize the rebalancing process of the imbalanced class distribution for further use in the feature selection procedure for dimensionality reduction for the medical X-ray image datasets. Different metaheuristic algorithms were used to maximize the parameter values of the rebalancing and feature selection phases to preprocess the datasets. The proposed work devised a multi-objective optimization strategy in the process of the metaheuristic algorithms search to solve the problem of dual imbalanced dataset and feature selection. Afterward, a comparative study of the proposed optimized approach with the conventional methods was conducted to evaluate the proposed method performance. The results established the superiority of the proposed method to overcome the imbalanced and multi-dimensional problem. The proposed method generated a reasonable number of minority class samples and selected a sensible subset of features to ultimately obtain a very extraordinary accuracy with great credibility from a negative value of kappa and a false high accuracy. It produced higher credibility and correctness classification performance in the practical problem of medical X-ray images compared to other algorithms. Feature selection with Random-SMOTE (RSMOTE) using the self-adaptive Bat algorithm is superior to the optimization using particle swarm optimization. The proposed method using the Bat algorithm achieved 94.6% classification accuracy with 0.883 Kappa value using the lung X-ray first dataset.