Reinforcement-Learning based Self-Healing Banking Transaction System against Malware Attacks in IoT Networks

Abstract

The IoT technology that was at the height of popularity in the banking industry has resulted, improved real-time transaction processing, enhancing the availability of smart financial products and services, and the automated payment systems are more feasible than the traditional banking IoT security models. However, these implementations often fail to detect newly developed "zero-day” malware applications. In contrast, the proposed design for this research is based on developing a Reinforcement Learning (RL) system for banking IoT networks. As compared to traditional Intrusion Detection Systems, which send an alert and/or block the offending actions, the RL-based systems described herein treat the network as a continuous agent-environment dynamic loop that autonomously configures the network, including real-time/configurable routing protocol and resource allocation, in order to preserve transaction integrity during active malware propagation. The proposed research presents a banking transaction processing system that employs RL technology to identify, mitigate, and recover from malware-related events and intrusions in IoT-enab led financial networks by using real-time virus/abnormality detection via a deep Reinforcement Learning (RL) controller. The deep RL controller models the overall context of network security in accordance with the principles of Markov Decision Processes (MDPs). The RL agent creates an effective mitigation strategy by detecting and isolating weak nodes, blocking harmful traffic, redirecting ways to conduct transactions, re-verifying identities safely, and reinstating service to the customer. The growing acceptance of Web of Things, or IoT, technologies in monetary infrastructures will enhance operational efficacy; however, it will also create new risks associated with connecting the banking industry to the Internet. Currently, the majority of malware detection systems based on machine learning and security products do not have any self-healing mechanisms for systems to recover from interruptions. This paper presents a new framework called the Reinforcement Learning-based Self-Healing Banking Transaction (RL-SHBT), which uses hybrid malware detection and enables flexible responses to attacks as well as automated methods for recovering from these attacks. The IoT banking environment has been modelled as a system based on Markov, where the use of a Deep Network with a set of flexible rewards will allow the DQN to learn the most effective response methods. This RL-based security model differs from existing models for the protection of IoT devices in that it utilizes transaction-sensitive state modelling and an adaptable reward system that encourages a balance between security and performance. The results of the experiments showed that this model has a detection accuracy of 99.02% (95% confidence interval: ±0.41%), a false positive rate of 1.12%, an average recovery time of 4.8 seconds, and a self-healing success rate of 97.8%. The research shows that an adaptive reward system with reinforcement learning speeds up the rate of learning by 34% compared to a fixed reward system. The results also indicate that reinforcement-learning-driven autonomous recovery greatly enhances cyber resilience and continuity of transactions within IoT-based financial systems. Machine learning and malware detection techniques have been designed to detect malicious network activities.