Optimized Epoch Selection Ensemble: Integrating Custom CNN and Fine-Tuned MobileNetV2 for Malimg Dataset Classification

In this paper, we introduce the Optimized Epoch Selection Ensemble, a new model designed to address key challenges in the application of deep learning within network security, specifically focusing on the effective selection of ensemble members. Our model combines a custom Convolutional Neural Network (CNN) with a fine-tuned MobileNetV2, targeting the classification of the Malimg dataset. By meticulously selecting the best-performing epochs from the CNN training and pairing them with the top epoch from MobileNetV2, we employ a strategy based on analyzing true predictions to optimize ensemble composition. This approach not only enhances the ensemble’s decision-making process but also significantly boosts accuracy, achieving 99.05% compared to the previous state-of-the-art result of 98.5%. Our results highlight the potential of tailored ensemble methods to significantly improve performance in specialized applications such as network security.