A Machine Learning-Based Approach for Detecting a Network Black Hole in Telecommunication Network

As network infrastructures become more complex, determining whether a failure of network management command is caused by a device malfunction or a network black hole has become a major challenge for Communication Service Providers (CSPs). In this paper, we propose a Network Black hole Detection (NBD) method to detect whether a network black hole exists in the telecommunication network. By analyzing historical incident reports and largescale network traffic data, we design two sets of features-entropy-based and ratio-based-that capture traffic variations among neighboring devices surrounding a suspected faulty node. We use machine learning algorithms such as XGBoost (XGB) and Logistic Regression (LR) to train predictive models based on these features. These models are then evaluated using real-world incident data collected from 208 access-layer Gigabit Ethernet Switch (GESW) devices. Experimental results show that, compared with the traditional rule-based method and the LR model, XGB model achieves the highest prediction accuracy ($66.7 \%$). In addition, the traditional rule-based method achieves $61.7 \%$ accuracy using our custom-designed features, which is significantly higher than the $50 \%$ accuracy from random guessing. Our experimental results demonstrate that the proposed NBD system can effectively distinguish between device failures and network black holes, offering practical value for network fault diagnosis.