Hidden Markov models for predicting cell-level mobile networks performance degradation

• Introduce a novel HMM-based framework for proactive cell degradation prediction in LTE networks. • Achieve high prediction accuracy (92.83%) using real-world operator data. • Demonstrate the potential for cost-effective mitigation and enhanced user experience through proactive interventions. • Provide a foundation for future research on integrating HMMs with Self-Organizing Networks (SONs) for autonomous network management. The ubiquitous deployment of Fourth-generation (4G) and Fifth-generation (5G) networks, while expanding user connectivity over a larger geographic area, presents significant challenges for mobile network management. The surge in connected devices increase cell congestion, while user mobility, fading, and interference further deteriorate channel quality. The combined effect of these factors contribute to a gradual degradation in cell and network performance, straining the capabilities of Mobile Network Operators (MNOs) to maintain optimal service quality. The degradation progresses through distinct stages, and ultimately leading to cell outages. Existing cell degradation detection methods, such as key performance indicators (KPIs) monitoring and drive tests, rely on indicators such as signal strength (like reference signal received power (RSRP)), network measurements (like traffic volume and handover requests), and user complaints. However, these methods may not provide real-time insights, leading to delays in identifying and addressing performance issues. This paper proposes a Hidden Markov Model (HMM) approach for proactive cell degradation prediction using real-time time-series KPIs data collected from a telecom operator’s network management system. The NetMax system, a network management system specifically designed for ZTE networks, was used for KPIs data collection. RSRP, 3GPP-recommended signal strength metrics for degradation measure, defines hidden states, while traffic volume and handover requests represent the observable states. The HMM achieves an average prediction accuracy of 93.12%, F1 score of 91.81%, and precision of 92.82% with 23 observation lengths, demonstrating the models’s effectiveness in predicting mobile network’s cell degradation severity.