Forgery detection in signatures is one of the major biological challenges in biometric authentication systems for the banking sector, as well as in many other legal and security domains. In this research, there will be a focus on the creation of a deep learning-based approach for identifying forged handwritten signatures using the powerful EfficientNet-B7 architecture integrated with an emphasis on improved feature extraction and discrimination through the convolutional block attention mechanism (CBAM). The general model is trained on a custom dataset containing genuine and forged signatures, taking into consideration preprocessing methodologies to improve the robustness of the network against noise and variability in writing styles. The CBAM mechanism enhances the network's attention toward focusing on what matters spatially and toward channels, thus improving the classification performance. The proposed model has produced near-perfect results of 98.33% in differentiating between real and forged signatures and 96.66% in authenticating genuine signatures, thus proving its efficacy and promise under real-world scenarios. Such results make our approach a giant leap in signature forgery detection, providing solutions that are supportive and scalable to strengthen security and authentication systems in documents. The work illustrates a promise for the integration of deep learning architecture and attention mechanisms toward detection of forgery and opens doors for further research in the area.