Towards Self-Adaptive Blockchain Security: An AI-Driven Framework for Resilient IoT Networks

¹ Research Scholar, Brainware University and Assistant Professor, MCA Dept, Techno International New Town, West Bengal, India
² Associate Professor, Dept of Computer Science and Engineering, Brainware University, West Bengal, India

^* Saswati Ghosh: This email address is being protected from spambots. You need JavaScript enabled to view it.

Abstract

The rapid growth of the Internet of Things (IoT) has rendered traditional static security mechanisms ineffective against increasingly complex and evolving cyber threats. While blockchain offers a decentralized and tamper-resistant approach to data management, it faces a fundamental challenge known as the “trilemma,” which involves balancing security, privacy, and scalability. This paper proposes an Intelligent Security Framework (ISF) as an advanced solution that goes beyond conventional cryptographic techniques.

The proposed architecture follows a defence-in-depth approach across multiple layers. At its foundation, it employs a hybrid cryptographic scheme combining AES-256 for efficient bulk data encryption and ECC for low-latency key management and authentication. To enhance privacy, the framework incorporates dynamic pseudonyms and metadata obfuscation to ensure anonymity and unlikability.

A key innovation of this work is the AI-driven Adaptive Consensus Engine, which analyses the behaviour of each block before validation. By integrating an LSTM-based neural network, the system can detect polymorphic threats, zero-day attacks, and abnormal transaction patterns in real time—capabilities that go beyond traditional signature-based methods. Additionally, a dynamically weighted Intelligent Reputation Score, powered by a Random Forest Classifier, is used to evaluate node reliability based on historical behaviour.

Experimental results demonstrate that the proposed framework improves detection rates by 18.4% compared to baseline hybrid models and achieves an anomaly detection accuracy of 98.5%. It introduces only a minimal processing overhead, with an average increase of 6ms in verification latency, while significantly reducing the false positive rate from 1.8% to 0.4%.