Sub-spatial prediction of votes integrating socioeconomic, educational, and age strata with machine learning and topological data analysis

The problem of election outcome prediction has been widely studied with various tools, and in recent years the study has grown considerably with methods taking spatial information into account. We present an innovative approach by integrating machine learning, topological data analysis, and geostatistics to predict voting preferences in the 2021 gubernatorial election in the Mexican state of Nuevo León, Mexico. In addition, we propose two classification models to predict voter preference, one geospatial and one non-geospatial, which result from evaluating multiple models. With these proposed models, we can predict the actual winner of that election. The geospatial model allows us to explore voting preferences by area, which could be of great importance to candidates in their electoral campaigns.