A Practical Study of Naive Bayes and Weighted Naive Bayes Cybersecurity Models for Wireless Networks
Oyinkansola Olagunju, Michael Abejide Adegoke, Gabriel Babatunde Iwasokun, Johnson Adeleke Adeyiga, Ojo Stephen Aderibigbe · American Journal of Networks and Communications · 2025
Intrusion Detection Systems (IDS) are essential for protecting wireless networks against an increasingly complex range of cyber threats. This study evaluates and compares the effectiveness of Naïve Bayes, Weighted Naïve Bayes, and Convolutional Neural Network (CNN) using the Network Security Laboratory-Knowledge Discovery and Data Mining (NSL-KDD) dataset. The Naïve Bayes model was used as a baseline due to its simplicity and computational efficiency, delivering consistent results across multiple iterations. The Weighted Naïve Bayes model, which incorporated a feature-weighting mechanism was used to improve precision and Receiver Operating Characteristics-Area Under Curve (ROC-AUC) scores while striking a balance between performance and interpretability. These qualities make it well-suited for real-time intrusion detection in wireless environments, where both transparency and resource efficiency are crucial. The Naïve Bayes model was used as a baseline for offering a straightforward and efficient classification approach with moderate overall performance while the Weighted Naïve Bayes model was used to enhance the standard version by introducing a feature-weighting mechanism, which improved precision and reduced false positives. The CNN model outperformed the Naïve Bayes-based approaches across all evaluation metrics, underscoring its ability to learn complex patterns in network traffic. The Weighted Naïve Bayes model was also used to strike a practical balance between accuracy and efficiency, making it especially suitable for wireless networks. The CNN model was used to deliver the highest scores across all evaluation metrics, including accuracy, precision, recall, F1-score, and ROC-AUC, reflecting its ability to learn complex patterns in the data. Experimental results demonstrated the potential of the Weighted Naïve Bayes model to support online learning and dynamic feature weighting, which is necessary for boosting adaptability to new and evolving attacks while preserving simplicity and transparency.