Methods, systems, and computer-readable storage media for automatic index creation for relational database systems. Query statements from a relational database are processed to generate query patterns from the query statements. Vectorization of the query patterns is performed to transform each query pattern into a numerical vector. A clustering algorithm is executed to cluster the numerical vectors into multiple clusters. Each cluster has a respective cluster center. A frequent query pattern is determined, for at least some of the multiple clusters, that corresponds to a respective cluster center. Active columns in the frequent query patterns are determined and a database index is automatically created for each active column that does not currently have a database index.