Multi-Task Learning for 5G Handover with Dynamic TTT Adjustment

In next-generation heterogeneous wireless networks (HetNets), ensuring stable connectivity amid dynamic user mobility remains a major challenge. Conventional handover strategies, such as the 3GPP Event A3 algorithm with fixed Time-To-Trigger (TTT) settings, often fall short in rapidly changing signal environments, leading to handover failures and ping-pong effects that degrade service quality. To address this issue, we propose a predictive, data-driven handover framework that combines Multi-Task Learning with Bi-directional Long Short-Term Memory (MTL-BiLSTM) networks and adaptive TTT control. Our method jointly predicts user trajectories and multiple Reference Signal Received Power (RSRP) values, enabling dynamic TTT adjustment based on statistical signal trends, including moving averages and standard deviation. The proposed approach adheres to 3GPP standards while offering enhanced responsiveness to real-time network fluctuations. Simulation results in realistic HetNet scenarios show significant improvements in handover efficiency, including reduced ping-pong occurrences and increased handover success rates. This work presents a scalable, intelligent mobility management solution for dense and complex 5G and beyond network deployments.