Accurately determining the gender of chickens holds significant benefits for poultry farms in optimizing breeding programs, improving productivity, and maintaining ideal gender ratios. In this study, we propose a novel approach using the EfficientNetB4 deep learning architecture, combined with customized fully connected layers and attention mechanisms, to enhance feature extraction for chicken gender determination. We trained and evaluated our model on a publicly available dataset containing 960 diverse chicken images, achieving superior performance over six state-of-the-art baseline models (AlexNet, GoogleNet, VGG-16, ResNet-18, DenseNet-201, and improved ResNet-50). Notably, our approach reached an accuracy, precision, recall, and F1 score of 98.75