Query optimization in database systems is a crucial issue and despite decades of research, it is still far from being solved. Nowadays, query optimizers usually provide hints to be able to steer the optimization on a query-by-query basis. However, setting the best-fitting optimizer hints is challenging. To tackle that, we present a learning-based approach to predict the best-fitting hints for each incoming query. In particular, our learning approach is based on simple gradient boosting, where we learn one model per query context for fine-grained predictions rather than a single global context-agnostic model as proposed in related work. We demonstrate the efficiency as well as effectiveness of our learning-based approach using the open-source database system PostgreSQL and show that our approach outperforms related work in that context.