Detecting Sleeping Cells in Cellular Networks Based on One-Class Support Vector Machines Algorithm and Deep Autoencoders

The internet's exponential growth has transformed industries, connected the world, and empowered individuals with unprecedented access to knowledge. Despite the widespread adoption of advanced mobile networks like 4G and 5G, regions like Africa and the Middle East still heavily rely on 3G technology. Self-Organizing Networks (SON) utilizing Minimization of Drive Tests (MDT) play a vital role in optimizing cellular networks. However, in Egypt, MDT is not utilized, leading to challenges in detecting and addressing sleeping cells-malfunctioning base stations that remain undetected, causing service disruptions. This research investigates previous methods for detecting sleeping cells and adapts them to suit Egypt's network characteristics. Two models, One-Class Support Vector Machines and Deep Autoencoders, are implemented, and their performance compared using data provided by Orange Egypt. The results demonstrate the effectiveness of Deep Autoencoders in detecting sleeping cells. This research contributes to improving network management and enhancing cellular network performance in Africa and the Middle East. Future work involves real-time deployment and extending the framework to include LTE KPIs.