EMPR-Balancer as an Oversampling Technique: A Case Study for Plant Disease Recognition

Class imbalance presents a key difficulty in machine learning, where the minority class appears in much fewer examples than the majority class. The presence of class imbalance often leads to biased models that fail to generalize well for the minority class. In this work, we propose EMPR-Balancer, a novel over-sampling technique based on Enhanced Multivariance Product Representation (EMPR). The proposed method aims to enhance the classification performance on imbalanced datasets by generating synthetic images for the minority class through EMPR's depth-wise components. We assess the performance of our approach alongside well-known oversampling methods on binary apple leaf disease classification. The results confirmed that the proposed method addresses class imbalance and improves both generalization and disease recognition performance.