Using machine learning (ML) approaches, this research presents a sophisticated strategy for HLWNets handover decision-making. A WiFi router is placed in one of the corners of a square atto-cell, which is surrounded by LiFi Access Points (APs) in the centre. In order to simulate user obstructions, a realistic object model was used, with an incidence rate of 1.5 per second. A user travelling at 3 m per second encounters obstacles as depicted by this model. Compared to regular versions, this one has a substantially higher blockage incidence rate. Using the XGBoost machine learning model, the article introduces a Higher Order (HO) method based on Reinforcement Learning (RL) that achieves a high accuracy of 98.5% in predicting user paths. At a user speed of 3 m/s, the RL-HO algorithm achieves an impressive 54% decrease in vertical handover (VHO) rates compared to standard LTE and 43% reduction in Smart HO methods, respectively. On top of that, as compared to the conventional methods, it increases the average throughput by a factor of 2.5. Results from the simulations prove that RL-HO can adapt on the fly to different user densities and speeds. As the number of users increases, it maintains constant throughput and makes appropriate handover decisions.