Analyzing the Performance and Stability of Cellular Networks -- Ermias Andargie Walelgne (Aalto University) -- Jukka Manner (Aalto Universtiy) -- Vaibhav Bajpai (Technical University of Munich) -- Joerg Ott (Technical University of Munich) The performance of a cellular network heavily depends on a number of factors; including limitation on hardware and software resources, radio technology type, quality of wireless links and network bandwidth, carrier networks, applications running on the devices, user mobility and time of the day. As a result, understanding and identifying key factors of cellular network performance that are representative of end-users devices and experience is still a challenging task. We use netradar, a measurement platform that measures and collects metrics related to cellular network performance collected from mobile users devices. Using this longitudinal dataset (around 4 years), we develop a methodology for understanding cellular network performance. We examined key characteristics of cellular network performance from the perspective of mobile user’s activity, mobile network operators, smartphone models, link stability, operating system platforms, location and time of the day. We perform a network-wide correlation and statistical analysis, to get a baseline understanding about individual network factors. We go further and use a machine learning approach for selecting important network features and to understand the relationship between various network factors. These features can then be used to build a prediction algorithm. Our results show that predicting network instability is possible using the minimal cellular network information, with up-to 90% of accuracy.