Amina Khacha

Amina Khacha

Auto-generated profile, based on publicly available information

Amina Khacha

LRSD Laboratory, Ferhat Abbas University of Setif 1, Setif, Algeria

Research Focus & Metrics

Academic Career Length

5Years (2022-Present)

(Calculated once Scilit publication data is synced: current year − year of first publication + 1)

Representative Publications

View All Publications
29 November 2022
Verified
Conference Paper
Closed access
AKAmina Khacha
Amina Khacha
RSRafika Saadouni
Rafika Saadouni
YHYasmine Harbi
Yasmine Harbi
ZAZibouda Aliouat
Zibouda Aliouat
The internet of things (IoT) is expected to offer a significant impact on the industry domain leading to the concept of industrial IoT (IIoT). The IIoT comprises machine-to-machine (M2M) and communication technologies with data automation and exchange to improve product quality and decrease pro-duction costs. As a consequence, a large amount of data is collected and smartly processed to provide optimal industrial operations. This growing deployment enables adversaries to con-duct potential and destructive cyber-attacks to accomplish their malicious goals. Therefore, intelligent decision-making actions for cyber-attack detection in IIoT are sorely required. To address this challenge, we propose an intrusion detection system (IDS) using deep learning models. Specifically, the proposed system is based on the combination of convolutional neural network (CNN) and long short-term memory (LSTM) that are excellent techniques for intrusion detection and classification due to their ability in classifying main characteristics and their effectiveness in performing faster computations. We adopt the most recent dataset named Edge-IIoTset that contains a real traffic network of IoT and IIoT applications. The proposed model is evaluated in terms of accuracy, precision, false positive rate, and detection cost within binary and multi-class classifications. The obtained results show that our CNN-LSTM model provides better performance and robustness in cyber security intrusion detection for IIoT applications compared to LSTM and traditional machine learning models. Moreover, it outperforms two recent related models in terms of accuracy rate.

13 February 2026 in IEEE Internet of Things Journal
Verified
Journal Article
Closed access
RSRafika Saadouni
Rafika Saadouni
CGChirihane Gherbi
Chirihane Gherbi
ZAZibouda Aliouat
Zibouda Aliouat
YHYasmine Harbi
Yasmine Harbi
AKAmina Khacha
Amina Khacha
HMHakim Mabed
Hakim Mabed
The integration of Internet of Things (IoT) technologies into smart agriculture has significantly enhanced automation, monitoring, and productivity. However, these systems introduce critical cybersecurity vulnerabilities and generate heterogeneous data, including structured network traffic and image-based inputs. This requires an intrusion detection system (IDS) capable of handling multimodal data, particularly since adversarial attacks can manipulate inputs to evade traditional detection models. To address these challenges, this article proposes a novel image-based IDS for smart agriculture environments. The system transforms network traffic into images and employs VGG16 for feature extraction, binary greylag goose optimization (BGGO) for feature selection, and a random forest (RF) for classification. It further integrates a dual defence strategy that combines a convolutional autoencoder denoising (CAED) module with adversarial training (AT) to improve robustness against adversarial perturbations. The proposed solution is evaluated on the CICIoT2023 dataset in eight traffic classes under three white-box adversarial attacks. The IDS demonstrates strong resilience across all perturbation levels. For weak perturbations ( $\varepsilon =0.01$ ), the dual defence achieves accuracies of at least 99.46%. Under moderate perturbations ( $\varepsilon =0.1$ ), it maintains high performance with macroaveraged accuracies of at least 99.39%. Even under strong perturbations ( $\varepsilon =0.3$ ), the system remains robust, attaining an accuracy of at least 96.50%. To assess generalization to real agricultural settings, the IDS is also tested using native crop images from the agricultural dataset. Under severe adversarial distortion ( $\varepsilon =0.3$ ), the system maintains robustness, achieving a macroaveraged accuracy of at least 96.71%. These results confirm that the proposed multimodal IDS provides a resilient, adaptive security solution for smart agriculture networks facing advanced adversarial threats.