A Decision-Tree-Based Algorithm for Proactive Handover Prediction in Multi-RAT Cellular Networks: A Drive-Test Study with Implications for 5G/6G Mobility Management

The combination of ultra-dense network deployments and high mobility results in an unfavorable outcome, rendering the task of handover more difficult than in environments typical of previous generations. 5G and 6G necessitate the deployment of heterogeneous networks and small cells to meet the demand, which at the same time introduces certain challenges. This scenario introduces small cells (such as femtocells, picocells, and microcells) that have very limited coverage areas, which, combined with the high speed of user equipment, create an excessive number of handover triggers, leading to the “ping-pong effect,” which wastes network resources and degrades the overall Quality of Service. Furthermore, high mobility means that a user might enter and exit a cell in less time than the mobile terminal’s dwell time, dropping the connection and resulting in handover failures and radio link failures. The conventional handover methods that rely on thresholds of certain factors such as the received signal strength could be insufficient for these environments. Different criteria should be balanced to avoid the drop, such as the user’s velocity, dwell time, target cell load, available bandwidth, device battery, and application latency requirements. Predictive methods could be a more efficient alternative to the existing reactive ones. This paper presents a decision-tree-based algorithm as one predictive method that learns the patterns among all the criteria mentioned and is particularly useful for avoiding ping-pongs and limiting handover failures. The classifier is trained on real multi-operator drive-test data with ping-pong events excluded from the positive class, and evaluated under Leave-One-Trace-Out cross-validation on 16 traces covering UMTS, HSUPA, HSPA+, and LTE cells. The proposed system achieves F1=0.642 and AUC =0.797 under LOTO, with a +0.052F1 lift over the best threshold-based baseline, while remaining interpretable and deployable in real time. The paper aims to present a solution applicable also to 5G NR and 6G.