Hybrid Transfer Learning and Deep Learning for Cloud Intrusion Detection Using Bio-Inspired Optimization
Salam Al-E’mari, Yousef K. Sanjalawe, Waleed Awwad, Hamzah Alqudah, Mohammad Adnan Aladaileh · 2025
This paper proposes a novel hybrid model for cloud intrusion detection, integrating transfer learning with bio-inspired optimization techniques to enhance detection accuracy and efficiency. Cloud environments present unique challenges due to their dynamic and shared-resource nature, making traditional Intrusion Detection Systems (IDS) inadequate. The proposed model leverages Convolutional Neural Networks (CNNs) for feature extraction, utilizing pre-trained models through transfer learning to reduce training time while maintaining high detection performance. In addition, bio-inspired optimization algorithms, specifically Cuckoo Search Algorithm (CSA) and Particle Swarm Optimization (PSO), are employed to fine-tune the CNN's hyperparameters, such as learning rate and batch size, ensuring optimal performance. The hybrid approach addresses the limitations of manual tuning and improves the model's ability to generalize across different cloud environments. Using the CICIDS 2017 dataset, the proposed system demonstrates superior performance in detecting both known and unknown threats, including zero-day attacks. The experimental results show significant improvements in accuracy of 99.3%, precision of 98.7%, recall of 98.5%, and an F1-score of 98.6% compared to state-of-the-art models.