Using machine learning for model benchmarking and forecasting of depletion-induced seismicity in the Groningen gas field