Spammer Detection On Computer Networks Using Gaussian Naïve Bayes Classifier And K-Medoids As Acquisition Training Data

Abstract

This research focuses on the implementation of the Gaussian Naïve Bayes algorithm for spammer detection in computer networks, leveraging K-Medoids clustering for training data acquisition. The increasing number of internet users, combined with the challenges of detecting spam activity on a network, has made manual detection ineffective. This study addresses the need for automated spam detection using machine learning algorithms. The Gaussian Naïve Bayes algorithm was chosen for its simplicity and effectiveness in handling continuous data, making it suitable for classifying network traffic as either normal or spammer. To acquire labeled training data, K-Medoids clustering was employed, offering robustness against outliers, which traditional clustering algorithms like K-Means often struggle with. The research involved collecting traffic data from a Mikrotik Routerboard at various intervals, followed by data preprocessing to remove irrelevant or null features. After preprocessing, the data was clustered using K-Medoids into two groups: spammer and normal. The Gaussian Naïve Bayes classifier was then applied to the clustered data, producing a model with high accuracy, precision, recall, and F1-score. Specifically, the model achieved 99.71% accuracy, 100% precision, 99.71% recall, and a 99.85% F1-score, indicating a well-balanced performance in spam detection. The results demonstrate that the Gaussian Naïve Bayes algorithm, combined with K-Medoids clustering, is effective for detecting spammers in computer networks. Future research could explore higher-layer network traffic and broader datasets, utilizing different routers for a more comprehensive evaluation. This approach provides a reliable solution for network administrators seeking to improve network security by detecting and mitigating spam activity.