Paper Details

Name: Reena
Country: India

Year: 2026
Volume: Volume-13, Issue-1 (January-June)
Page Number: 30-39
DOI: https://doi.org/10.5281/zenodo.18654496

ABSTRACT
The high rate of Internet of Things (IoT) technologies and networks development has greatly exposed businesses to advanced cyber-attacks, especially zero-day attacks that use the vulnerabilities that were unfamiliar before. Because they rely on established patterns, standard signature-based intrusion detection systems (IDS) are unable to identify such attacks. This paper presents a strong deep learning-based intrusion detection model which uses the autoencoder designs to detect network behavior that is abnormal. The model can effectively detect deviation of normal traffic that can be viewed as a symptom of a zero-day attacks by learning representations of normal traffic, but the model keeps the false-negative
rates at a minimum. Evaluation of the suggested approach is performed with the help of benchmark datasets, i.e. NSL-KDD, CICIDS2017, and IoT-based traffic gathering and contrasted with conventional machine learning techniques like One-Class Support Vector Machines. The results of the experiment prove that the models using autoencoders have better
detection accuracy, recall and F1-scores, especially in complicated and low-volume attack conditions. The results confirm that deep autoencoders are an appropriate choice in scalable, adaptive, and high-performance zero-day intrusion detection in contemporary IoT and cyberphysical network settings.

Keywords: Zero-Day attacks; Intrusion Detection Systems (IDS); Deep learning; Autoencoders; Anomaly Detection; machine learning.