Anomaly based Intrusion Detection to Enhance IoT Devices Security using LSTM and Radial Basis Function Approach

The growing prevalence of IoT devices has rendered their security a paramount issue. Intrusion detection is essential for protecting IoT systems from potential attackers. Anomaly-based intrusion detection systems, despite their capacity to identify previously undiscovered threats, are frequently dis...

Ausführliche Beschreibung

Gespeichert in:

Bibliographische Detailangaben
Veröffentlicht in:	2025 3rd International Conference on Data Science and Information System (ICDSIS) S. 1 - 6
Hauptverfasser:	Narayana, G. V. L, G, Sharmila, A, Mohaideen, Sathya, S., Vennila, C., Nishant, Neerav
Format:	Tagungsbericht
Sprache:	Englisch
Veröffentlicht:	IEEE 16.05.2025
Schlagworte:	Anomaly detection distributed convolutional neural network (DCNN) Faces Feature extraction improved conditional variational autoencoder (ICVAE) Information systems Internet of Things Long short term memory long short-term memory (LSTM) Network intrusion detection network intrusion detection systems (nids) Principal component analysis principal component analysis (PCA) Security Training
Online-Zugang:	Volltext
Tags:	Tag hinzufügen Keine Tags, Fügen Sie den ersten Tag hinzu!

Beschreibung
Zusammenfassung:	The growing prevalence of IoT devices has rendered their security a paramount issue. Intrusion detection is essential for protecting IoT systems from potential attackers. Anomaly-based intrusion detection systems, despite their capacity to identify previously undiscovered threats, are frequently disregarded in favour of signature-based detection methods. Nonetheless, anomaly-based methods can be equally efficacious in thwarting intrusions. A significant problem in this field is assessing the efficacy of diverse anomaly detection strategies, which has led to reluctance among organisations in implementing these methods. The suggested methodology comprises three fundamental steps: data cleansing, feature extraction, and model training. During preprocessing, the most pertinent properties are chosen. PCA is utilised for feature extraction to discern the most informative attributes. The model is further trained with LSTM-RBF networks to improve detecting skills. The experimental findings demonstrate that the suggested method surpasses the two predominant alternatives, LSTM and RBF, attaining a detection rate of 97% and a low false alarm rate of 2%. The findings underscore the exceptional efficacy of the proposed LSTM-RBF method in accurately detecting intrusions while minimising false alarms, thereby illustrating its potential to enhance IoT security frameworks and promote the broader implementation of anomaly-based techniques in practical applications.
DOI:	10.1109/ICDSIS65355.2025.11071203