Schools and school population are prone to safety and health risks due to their proximity to hazardous features, either natural or infrastructural. In Mexico there is an official standard that points out which features to consider when selecting the location for building a school, and how proximal these features can be to the school’s location in order to not represent a threat. This work focused on applying both geospatial analysis and unsupervised machine learning techniques to detect the hazardous features per school in Mexico, and group these schools as per the data patterns themselves. For this, a data set containing Mexico’s schools and the proximal hazardous features for each school was built by spatially combining multiple official data sets. After that, the K-Modes partitional clustering machine learning algorithm was used with the created dataset. Multiple clustering models were built with this algorithm by testing various K values (number of clusters), and their clustering quality was measured with internal clustering evaluation metrics. The clustering model with the highest quality was the one that grouped Mexico’s schools into 11 clusters, each one indicating the most common(s) hazardous feature(s) to the schools of each cluster. The evaluation metrics results for this model were: Silhouette Score (0.72), Calinski-Harabasz Index (32863.96), Davies-Bouldin Index (0.72) indicating strong clustering results, with well-separated and cohesive clusters. The study was carried out by following the Cross Industry Standard Process for Data Mining (CRISP-DM) methodology, consisting in multiple phases and tasks for data mining projects execution.