Performance of Machine Learning-Based Multi-Model Voting Ensemble Methods for Network Threat Detection in Agriculture...

Area of Science:

• Agricultural Technology
• Cybersecurity
• Machine Learning

Background:

• Agriculture 4.0 integrates Information and Communication Technologies (ICT) into farming operations.
• This integration introduces significant cyber threats, necessitating robust security measures.
• Network traffic analysis and classification are crucial for mitigating these threats.

Purpose of the Study:

• To evaluate the effectiveness of various Machine Learning (ML) classifiers for network traffic classification in the context of Agriculture 4.0.
• To compare the performance of individual ML models against ensemble models.
• To assess performance across different dataset variations (initial, undersampled, oversampled NSL-KDD).

Main Methods:

• Implementation and evaluation of individual ML classifiers: K-Nearest Neighbors (KNN), Support Vector Classification (SVC), Decision Tree (DT), Random Forest (RF), and Stochastic Gradient Descent (SGD).
• Development and assessment of ensemble models: hard voting and soft voting classifiers.
• Utilized three variations of the NSL-KDD dataset for comprehensive performance analysis.

Main Results:

• Individual ML algorithms demonstrated varying performance levels across dataset variations.
• Ensemble models (hard and soft voting) generally achieved higher accuracy compared to individual ML classifiers.
• The performance gains of ensemble methods were observed consistently across the tested dataset variations.

Conclusions:

• Ensemble ML models offer superior network traffic classification accuracy for securing Agriculture 4.0 environments.
• The proposed approach provides a viable solution for identifying and mitigating cyber threats in smart farming.
• Further research can explore advanced ensemble techniques and real-world deployment scenarios.