The financial sector is grappling with significant challenges in detecting cyber attacks, leading to potential short- and long-term financial losses for banks and other institutions. The statistical and machine learning methods have been effective in identifying suspicious activities, they have struggled to achieve a balance between recall and precision. To improve accuracy, this paper introduces a novel approach that employs deep learning and bio-inspired algorithms to detect suspicious activities. The proposed model analyzes transactional patterns, quantities, and temporal aspects using a carefully curated dataset of labeled transactions. The model shows promising results in distinguishing between legitimate and fraudulent operations, achieving a balance between recall and precision. Further, many in the industry are transitioning to cloud computing infrastructures to enhance application performance. However, these infrastructures are not ideal for delay-sensitive applications, such as those in the medical and finance sectors. To address communication delays, fog computing has emerged as a new paradigm. The  proposed model was simulated using Python and the Google Colab framework, and  experimental results shows that  improved accuracy and a balanced recall and precision.