This study presents a machine learning-based RL-HO method for handover decision-making in HLWNets, combining XGBoost and reinforcement learning. Simulations with users moving at 3 m/s and high blockage incidence achieved 98.5% path prediction accuracy, reduced vertical handover rates by 54% versus LTE and 43% versus Smart HO, and increased average throughput by 2.5×. RL-HO adapts to varying user densities and speeds while maintaining performance.