Ensemble Model of Convolutional Neural Networks Based on Model Averaging

Model averaging is essential for accurately classifying diverse samples. This study integrates multiple convolutional neural networks (CNNs) to harness their complementary strengths and optimize classification performance on the CIFAR-10 and CIFAR-100 datasets. Evaluated by accuracy, recall, and F1 score, the proposed approach yields notable improvements, including accuracy increases of 0.3 for the ‘automobile’ and ‘ship’ classes and 0.6 for the ‘frog’ class, along with enhanced metrics across other categories.