Knowledge of land values is crucial for the development of economic and fiscal policies that enable the estimation of comprehensive property values in alignment with the dynamic nature of the real estate market. In Colombia, the cadastre has undergone a significant transformation, evolving from a purely fiscal model to a multipurpose cadastre aimed at addressing contemporary challenges such as outdated records, incomplete coverage, and the lack of systematic training. Consequently, continuous cadastral updating and professional training are essential to accurately reflect economic development, urban expansion, and administrative changes. However, despite advancements in technology, the available tools have not been sufficient to efficiently capture, process, and analyze the vast volume of information required for an accurate valuation system. This paper presents a theoretical approach to evaluating and selecting an artificial intelligence-based methodology to improve urban land valuation. The machine learning models analyzed include Random Forest (RF), Support Vector Machine (SVM), Artificial Neural Network (ANN), and Gradient Boosting Model (GB). Ultimately, the study highlights the effectiveness of Random Forest (RF), which, when integrated with the Kriging geostatistical model (KG) to incorporate spatial dependencies, achieves a superior predictive accuracy.