Modelo de aprendizaje de máquina para la verificación de la calidad de los datos en actualizaciones catastrales en municipios de Casanare

Abstract

In the context of cadastral updates in Colombia and their importance for governmental decision-making, data quality plays a crucial role in the proper development of cadastral processes. It is necessary to streamline these processes, make them more efficient, and avoid inconsistencies as defined in the quality models set by the Agustín Codazzi Geographic Institute (IGAC) to advance in these processes. The application of technologies such as machine learning helps increase efficiency in these procedures. As a solution, a machine learning model was proposed to verify the quality of cadastral data in municipalities of the Casanare department through the definition of four quality levels, and the selection of variables that represent the territorial reality. In total, 20 variables were selected, among which the most influential were land area (alphanumeric and digital), property registration number, type of domain, and economic purpose. The variables with the most inconsistencies were registry area, cadastral area, complete name information, and records of legal and natural persons, as well as the discrepancy between the registered area and the calculated area. The four levels are divided into the verification of the completeness of the information, the structure of the national cadastral number, the compliance with consistency rules, and topological validations. Of the evaluated property data, 70% meets the quality parameters, which reveals a need to implement new processes, like the one presented, for more adequate data management and recording. Additionally, the model's performance had an RMSE of 0.0317, an MAE of 0.0185, and an R² Score of 0.6629, indicating good performance, closeness of predicted values to real values, and variability in the data with room for improvement.