CBFF-LSTM: a deep learning framework for early-stage DOS attack detection in IOT-enabled WSN through correlated behavioural feature fusion

Denial-of-Service (DoS) attacks pose significant threats to Internet of Things-enabled Wireless Sensor Networks (IoT-WSN) due to device vulnerabilities and resource constraints, leading to network disruptions and service degradation. Existing approaches primarily focus on post-attack detection and classification. This paper proposes a novel deep learning framework called Correlated Behavioural Feature Fusion with Long Short-Term Memory (CBFF-LSTM) for proactive early-stage DoS attack detection with minimal computational overhead in IoT-WSN environments. The CBFF-LSTM framework introduces three key innovations (1) a progressive four-stage detection mechanism enabling early attack identification, (2) a novel Correlated Behavioral Feature (CBF) extraction methodology that simultaneously analyzes spatial–temporal dependencies at both node and network levels, and (3) an adaptive Feature Fusion Vector (FFV) generator optimized for resource-constrained IoT devices. Experimental validation was conducted using NS-3 simulator, incorporating diverse attack scenarios. A performance with 98.4% detection accuracy under controlled simulation (realistic expectation: 94–96%), 1.4% false positive rate (expected real-world: 2–4%), and early detection (2.3 s average) was achieved. The framework achieves linear scalability O(n) with inference times of 1.8 ms per sample and resource efficiency below 6% CPU usage, making it suitable for resource-constrained IoT environments while providing predictive capabilities not achieved by existing LSTM and feature fusion combinations.