ECLNet - A Lightweight CNN-LSTM Fusion Network with Adaptive Feature Pruning for Real-Time Intrusion Detection in Edge-IoT Environments

ECLNet is a new lightweight deep learning model designed especially for real-time intrusion detection in Edge-IoT settings. Through the combination of Convolutional Neural Networks (CNN) and Long Short-Term Memory (LSTM) networks, it can well identify spatiotemporal features from network traffic data. To maximize efficiency, ECLNet uses an Adaptive Feature Pruning (AFP) algorithm that cuts down latency and memory consumption by eliminating redundant or less significant features. The model provides 99.2% detection accuracy on the CICIDS2017 and BoT-IoT datasets, which indicates its ability in intrusion detection. In addition, it provides 45% time reduction in inference and 32% memory reduction compared to conventional deep learning approaches, which makes it appropriate for deployment on limited-resource IoT devices. In conclusion, ECLNet provides an efficient and practical solution to improve security in real-time edge computing settings.