Mobile Networks (MNs) are rapidly evolving, enabling seamless and faster communication around the world. However, the escalating complexity and management challenges of modern MNs, have become more pronounced, especially with the advent of 5G technology. User throughput is a pivotal determinant of Quality of Service (QoS) in MNs, influenced by many dynamic factors in such environments. Conventional monitoring tools often encounter limitations in comprehensively capturing the evolving nature of these networks, impeding the effective identification of some throughput constraints. To address this challenge, this work introduces a data-driven approach for unsupervised anomaly detection in MNs, specifically tailored for detecting anomalous throughput cells. This involves developing a scoring methodology leveraging Performance Management (PM) data. The proposed method integrates Machine Learning (ML), eXplainable Artificial Intelligence (XAI), and statistical analysis to evaluate the extent to which each cell’s user throughput aligns with the expected performance for its conditions, relative to comparable peers. This enables the identification and scoring of local anomalies where throughput falls relatively below expectations. Validation of this approach on a real-world dataset has demonstrated its effectiveness in identifying cells with the potential for throughput enhancement. This can help Mobile Network Operators (MNOs) improve operational efficiency and provide a better customer experience.