A Scalable Hybrid Autoencoder–Extreme Learning Machine Framework for Adaptive Intrusion Detection in High-Dimensional Networks

Anubhav Kumar, Rajamani Radhakrishnan, Mani Sumithra, Prabu Kaliyaperumal, Balamurugan Balusamy, Francesco Benedetto

The rapid expansion of network environments has introduced significant cybersecurity challenges, particularly in handling high-dimensional traffic and detecting sophisticated threats. This study presents a novel, scalable Hybrid Autoencoder–Extreme Learning Machine (AE–ELM) framework for Intrusion Detection Systems (IDS), specifically designed to operate effectively in dynamic, cloud-supported IoT environments. The scientific novelty lies in the integration of an Autoencoder for deep feature compression with an Extreme Learning Machine for rapid and accurate classification, enhanced through adaptive thresholding techniques. Evaluated on the CSE-CIC-IDS2018 dataset, the proposed method demonstrates a high detection accuracy of 98.52%, outperforming conventional models in terms of precision, recall, and scalability. Additionally, the framework exhibits strong adaptability to emerging threats and reduced computational overhead, making it a practical solution for real-time, scalable IDS in next-generation network infrastructures.