This paper presents an update to our previous research on predicting shear wave velocity (Vs) from piezocone penetration tests (CPTu) using machine learning models. In our prior work (Entezari et al. 2022), we leveraged a substantial dataset of over 100,000 paired Vs-CPTu records from seismic piezocone (SCPTu) soundings across diverse soil types and geological settings. Our initial study demonstrated that our machine learning models outperformed some of the widely-used CPT-based Vs prediction methods (e.g. Mayne 2006, Robertson 2009). In this follow-up study, we investigated the impact of geographical information, specifically latitude and longitude, on our Vs prediction models. Three new models were developed for all-soils, uncemented, and uncemented soils using the basic CPTu data including depth, corrected tip resistance (qt), sleeve friction (fs), dynamic porewater pressure (u2) as well as longitude, and latitude as input parameters. The performance of the models was compared to the ones developed using only basic CPTu data. The results revealed a noticeable improvement in prediction accuracy when incorporating spatial context. This additional layer of data enhances the robustness and precision of our machine learning models. This improvement is potentially attributed to the influence of geological variations associated with geographical location on shear wave velocity. Furthermore, to gain insights into the importance of different features and variables in our models, SHAP (SHapley Additive exPlanations) analysis was employed, allowing to assess the contributions of CPTu parameters and geographical information in the predicted results and model performance. The inclusion of geographical data appears to enhance the performance of machine learning models in predicting Vs from CPTu tests. This approach shows significant potential for geotechnical applications, emphasizing the importance of spatial context in soil property assessments. As we continue to advance our understanding of soil behavior, the incorporation of geospatial information into machine learning models offers a compelling avenue for improving geotechnical predictions.