The Effectiveness of Data Augmentation in Image Classification Using Whale Optimization Algorithm

Imbalanced datasets significantly degrade the performance of image classification systems, particularly in application domains where data collection is difficult, such as medical diagnosis and plant disease analysis. Data augmentation is a commonly used strategy to alleviate this problem by increasing the number of samples in minority classes. In this study, a novel image augmentation approach based on optimized edge detection, named Whale Optimization Algorithm–based Image Augmentation (WOA-IA), is proposed. Unlike conventional augmentation techniques, the proposed method generates synthetic samples by combining original images with optimally extracted edge representations. Edge detection is performed using a multi-layer filter structure whose parameters are optimized via the whale optimization algorithm by maximizing similarity with reference edge images. The resulting edge images are then integrated into an angular placement-based augmentation model. The effectiveness of the proposed WOA-IA method is evaluated on five widely used imbalanced datasets—Plant Leaves, Flowers102, Cassava, Plant Village, and HAM10000—using pre-trained CNN models including DenseNet201, MobileNet, and VGG16. Experimental results demonstrate that the proposed method consistently improves classification performance compared to both imbalanced datasets and standard edge-based augmentation techniques, particularly for datasets with high imbalance ratios.