Human Activity Recognition (HAR) is a rapidly evolving research area that focuses on identifying and classifying human activities using sensor- and vision-based data. HAR has gained significant attention due to its wide range of applications in intelligent surveillance systems, biometric identification, smart environments, healthcare monitoring, and human–computer interaction. Accurate and real-time activity recognition is particularly important in surveillance applications, where the timely detection of suspicious behaviors can contribute to crime prevention and public safety. Recent advances in deep learning have demonstrated remarkable performance in HAR tasks, especially through Convolutional Neural Networks (CNNs), which are capable of automatically extracting discriminative features from visual data. However, the performance of CNN-based models is highly dependent on the selection of optimal network parameters. To address this challenge, this paper proposes a hybrid HAR framework that integrates CNNs with the Bat Optimization Algorithm (BOA) to enhance feature learning and classification performance. The proposed method is evaluated using the Weizmann human activity dataset and compared with several existing HAR approaches. Experimental results demonstrate that the integration of BOA with CNN improves recognition accuracy and classification effectiveness, highlighting the potential of the proposed framework for intelligent activity recognition applications.