IoT Networks
Khushwant Singh, Mohit Kumar Yadav, Yudhvir Singh, Pratap Singh Malik, Vikas Siwach, Daksh Khurana, Binesh Kumar, Ramesh Kumar Yadav, Ahmed A. Elngar · 2024
Financial fraud is a persistent problem for consumers and financial institutions worldwide. It loses billions of dollars annually. Consequently, a strong fraud detection system (FDS) is essential to minimizing damage to financial institutions as well as clients. One common technique for spotting fraud is to use machine learning algorithms, which analyze large volumes of data to help with pattern detection and future prediction. It is difficult for a centralized FDS to detect fraud trends when these problems are coupled. To train a fraud detection model, this work presents a framework for federated learning, a machine-learning environment in which several entities collaborate to solve a machine-learning problem under the guidance of a central server or service provider. Also, the chapter examines how combined learning can be used to protect privacy in machine learning in Internet of Things systems. It focuses on four main calculations: federated averaging (FedAvg), secure aggregation, holomorphic encryption-based federated learning, and differential privacy in combined learning. Extensive experiments were carried out to evaluate these computations in terms of proving accuracy, conserving protection, and computing efficiency. The findings are shown in the results, with FedAvg achieving the highest accuracy of 92.5% and secure conglomeration demonstrating competitive precision levels of 91.8%. Calculations for differential privacy and holomorphic encryption demonstrated strong security conservation with very little data leakage and security parameters of 2.5 and 1.0, respectively. With little communication overhead and the ability to alter accuracy and conserve protection, secure aggregation emerged as a potential configuration. The computational productivity assessments revealed that secure accumulation produced little communication overhead despite its strong security conservation, which makes it suitable for IoT scenarios with limited resources. By using this tactic, financial institutions may avoid sharing datasets and benefit from a shared model that has seen more fraud than any one bank has on its own. Thus, the sensitive data of the user is protected. The results of the chapter indicate that the federated model (federated averaging) may be as good as or better than the central model (multi-layer perceptron) in detecting financial fraud. This chapter adds to the growing conversation around mixed learning in the Internet of Things by providing insights into the trade-offs between accuracy, security, and efficacy and by laying the groundwork for future developments in privacy-preserving machine learning standards.