Advancing Intrusion Detection Using Deep Learning: A Hybrid Approach

Title

Advancing Intrusion Detection Using Deep Learning: A Hybrid Approach

Creator

Bellary, Keertipriya; Jalapur, Shruti; Diana Jeba Jingle, I.

Description

Intrusion detection systems (IDSs) are vital for securing networks against evolving cyberthreats. Traditional machine learning models often struggle with complex network traffic and imbalanced attack patterns. This study proposes an advanced ensemble model integrating ANN, LSTM, random forest, and LightGBM to enhance detection accuracy and robustness. Evaluations on the KDD99 dataset demonstrate that the ensemble model outperforms standalone ANN-LSTM models, achieving 92.4% accuracy, 97.4% precision, 87.1% recall, and a 91.9% F1 score. Hybrid models also showed significant improvements, with Nadam optimization yielding an F1 score of 93.10% for ANN-LSTM-random forest and Adam optimization achieving 93.30% for ANN-LSTM-LightGBM. By addressing data imbalance and improving attack pattern detection, this approach provides a scalable, efficient solution for real-time intrusion detection with superior performance. The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2026.

Source

Lecture Notes in Networks and Systems;Volume;1614 LNNS;pp.251-264

Date

01-01-2026

Publisher

Springer Science and Business Media Deutschland GmbH

Subject

Accuracy; Artificial neural networks (ANNs); Cybersecurity; Ensemble model; F₁ score; Hybrid models; Intrusion detection system (IDS); KDD99 dataset; LightGBM; Long short-term memory (LSTM); Nadam); Optimizers (Adam; Precision; Random forest; Real-time detection; Recall

Coverage

Bellary K., Department of Computer Science and Engineering, Christ University, Bangalore, India; Jalapur S., Christ University, Bangalore, India; Diana Jeba Jingle I., Christ University, Bangalore, India

Rights

Restricted Access; Hardcopy may be available in the library

Relation

ISSN: 23673370; ISBN: 978-981952877-6;

Format

online

Language

English

Type

Conference paper

Identifier

https://doi.org/10.1007/978-981-95-2878-3_18

https://www.scopus.com/pages/publications/105029586132?origin=resultslist