Publication Date

Spring 2026

Degree Type

Master's Project

Degree Name

Master of Science in Computer Science (MSCS)

Department

Computer Science

First Advisor

Navrati Saxena

Second Advisor

Thomas Austin

Third Advisor

Pooja Shyamsundar

Keywords

Unstable handover, supervised machine learning, 5G/LTE, offline gating, sensitivity analysis

Abstract

Frequent and unnecessary handovers are a major challenge in dense cellular networks because they increase signaling overhead and reduce the benefit of mobility management. This project studies unstable handover prediction using a machine learning pipeline with traces extracted from ns-3 LTE simulation and SUMO mobility. We define the main target as unstable 15 = reversal 15 OR short dwell 15. We then evaluate ML models like Logistic Regression, Explainable Boosting Machine (EBM), LightGBM, and XGBoost on dataset grouped by train-test splits with respect to UE and performing offline gating analysis. Results show that boosted-tree models perform better than the linear baseline on ranking metrics, and LightGBM has the best balanced threshold tradeoff for offline gating. The study also shows that pre-handover measurements can be used to find a large amount of unstable handover events with some stable handover loss. It also highlights the difficulty of the overall task and how model and corresponding threshold choice affect the performance of offline gating.

Download

Available for download on Saturday, May 22, 2027

Share

COinS