Designing a Convolutional Neural Network (CNN) topology with optimal performance is a challenge. This paper proposes a hybrid algorithm combining the nature-inspired Bees Algorithm (BA) with Bayesian Optimization (BO) technique to improve CNN performance (BA-BO-CNN). In addition, another hybrid algorithm is proposed which uses BA to optimize CNN hyperparameters (BA-CNN) to improve the network performance. Applying the hybrid BA-BO-CNN rather than BA-CNN on human electrocardiogram (ECG) signals the testing accuracy improved from 92.50% to 95%, on Cifar10DataDir benchmark data the accuracy on the validation set increased from 80.72% for BO-CNN to 82.22% for BA-BO-CNN, and finally, on benchmark digits images the training, validation and testing accuracies remained the same compared to the existing BO-CNN, but with more efficient computational time since it is reduced by 3 minutes and 12 seconds for BA-BO-CNN and 4 minutes and 14 seconds for BA-CNN.