Establishing an efficient security model using learning and optimization approaches

doi:10.1007/s10586-025-05914-9

2026-01-12

PRE

DOI：10.1007/s10586-025-05914-9

原文链接

原文求助

摘要

En 中文

Risks related to cybersecurity, like cyberattacks through smart devices, are escalating in frequency and severity in recent times. The wide-ranging data traffic in communication within the devices connected to the Internet of Things (IoT) poses a substantial problem in protecting the devices from possible breaches in security, mainly when network traffic data is unbalanced. To address this issue, we introduce a hybrid Dense Convolutional Neural Model with Orca Predation Algorithm (DCNM-OPA) combined with ensemble classifiers (k-NN and Random Forest) for accurate intrusion detection. The framework involves Z-score normalization for preprocessing, DCNM with soft thresholding for effective feature extraction, and OPA for hyperparameter tuning and dimensionality reduction. In the presented work, as the first step, DL algorithms are employed to obtain vital features accurately from a BoT-IoT dataset, which offers realistic network traffic. Then, the malware-detecting efficiency of ten various ML models was evaluated. We also analyze our model using k-NN and Random Forest ensemble classifiers and Dense convolutional network model (DCNM) DL architectures. The offered research work delivers an innovative method that associates optimization, deep learning and ensemble learning methods to identify cyber-attacks in the Internet of Things environment accurately. The input data is pre-processed to convert it into an appropriate format. Then, the Orca Predation Algorithm (OPA) is applied to decrease the complexity of the high-dimensional data, and the relevant features are selected. The cyber-attacks are identified and classified through training the proposed network model with the ensemble classifier. The proposed network model’s performance is fine-tuned by applying the OPA algorithms. The projected approach aims to enhance the IoT environment’s security by distinguishing and categorizing intrusions effectively. Investigational assessments performed on benchmark datasets, mainly the Internet of Things dataset, prove that the projected OPA with the Dense Convolutional Network model attains 99% average accuracy. Likewise, quantitative performance outcomes have been compiled to highlight the significance of our model. Specifically, the DCNM-OPA achieves 99.6% accuracy, 98.3% precision, and 99.3% F1-score on benchmark datasets (BoT-IoT and UNSW-NB15), surpassing the baseline CNN, LSTM, and traditional ML methods.

Keyword:

Network

Attack

Prediction

Deep learning

Optimization