Volume 16, No 1, 2019

Building Explainable AI Systems With Federated Learning On The Cloud


Vijaya Venkata Sri Rama Bhaskar , Pradeep Etikani , Krishnateja Shiva , Ashok Choppadandi , Arth Dave

Abstract

The extensive usage of data processing-based systems and services—many of which rely on Artificial Intelligence (AI) and, more precisely, Machine Learning (ML) algorithms—defines the present day. Detecting fraudulent transactions continues to be a major challenge for financial institutions worldwide. The banking sector must have sophisticated fraud detection systems in place to protect their assets and maintain consumer confidence, but there are a few obstacles that must be overcome in order for such systems to be developed effectively and efficiently. Sensor networks, which constitute the basis of the Internet of Things, are employed in applications related to safety, healthcare, and the military. Threats to the safety of the Internet of Things-based Wireless Sensor Networks (IoT-WSNs) can come from a variety of sources. This research presents safe attack localisation and detection in IoT-WSNs to improve security and the provision of services. Before beacons nodes broadcasted data to the base location, the approach generated block chain trust values using a hierarchical architecture based on block network-based cascade encrypting and trust assessment. Furthermore, deep learning systems lack the explain ability of the projected outcomes, which is often needed in medical health. This makes them similar to "black boxes." This restricts how deep learning may be used in actual healthcare systems. Malicious nodes are classified using federated learning. Federated learning combines methods such hybrid random forests, gradients boost, collective wisdom, K mean clustering, and guidance vector machine learning with a feature evaluation process to classify risky nodes. Compared with the current approaches, average recognition and categorization accuracy of the proposed system is 100% for binary and 98.95% for multiple classes. This demonstrates the effectiveness of the suggested approach for large-scale IoT-WSNs in terms of both performance and security. The suggested approach leverages heterogeneous wireless sensor networks to provide secured services.


Pages: 371-385

Keywords: IoT-WSNs, Artificial Intelligence (AI), Fraudulent Transactions, Federated Learning, Machine Learning (ML), Deep Learning Models, Malicious Nodes, Classification Accuracy, Blockchain-Based, Fraud Detection Systems, Detection Systems.

Full Text