Integrating Enhanced Optimization Techniques with Lightweight Cryptographic Algorithms for Robust Attack Detection and Data Protection
P Sai Soujanya, P. Salini · 2024
IoT devices have seen an increase recently, many of which are resource constrained and capable to provide vulnerabilities that could be exploited through a wide range of cyberattacks. Sandboxing is one way in order for the system security level to become more effective with respect how it can handle specific threats as well reduce its response time where traditional complex antimalware chain reaction signatures speed was never that solid giving new ground literals on quick granular boastful approach when performing proper access control. Therefore, this work introduces an integrated Particle Swarm Optimization (PSO) algorithm with Lightweight Homomorphic Cryptography based approach to improve attack detection capability and data security for Data Stores in IoT enabled environments. A PSO algorithm is used to iterate over the Hyper-parameters in Hybrid Convolutional Autoencoder for Real-Time Network Attack Detection. Our experimental results indeed show a considerable elevation in the detection accuracy across different types of attacks including DDoS, Man-in-the-Middle, Data Exfiltration and Brute Force attacks. The detection average of the optimized HCAE model is 98.5% and for DDoS attack 99.7% was registered, being this attacked classified in highest accurate one detected by our proposal seen at Table (3). This is an increase of 7.4–12 % compared to pre-optimization adversarial accuracy for various attack types [68]. It reduces the false positive rate to 1.2%, so It increases reliability of Detection system also. According to computational efficiency, the system shows that the CPU usage drops from 75% to 55%, Memory from Drops out of memory too(i.e.80MB - >65MB) The reduction in detection latency is 37% which changes from a set of 150ms to only about 95ms, so it enables the live deployment along with real-time. The LHC scheme is equally vital for protecting data from resource-limited IoT devices, and provides an intrinsic overhead to the processing of data that tends toward zero with encryption/decryption operations taking on average only 5ms. All while able to scale up with data size up so that it's actually useful at real IoT large and medium-size deployments. We also validate the adaptability of our system to changes in network topology and new devices or removal. Comparing the results, we can observe that accuracy in one class detection remains relatively constant over time with some variation reflecting the reflective of dynamicity of IoT environments. Its detection rate is higher than signature-based intrusion de- tection systems and conventional cryptographic methods with a lower false positive rate, in addition to being computationally efficient which renders it suitable for IoT security.