Spectrum occupancy prediction allows cognitive radio secondary users to exploit temporal spectrum opportunities one step-ahead. Temporal correlations in spectrum sensing measurements can be utilized to predict primary user activity patterns. Where applicable, cooperative spectrum prediction has the potential to improve prediction accuracy compared to single user (local) spectrum prediction. This letter presents the concept and methods for soft fusion-based cooperative spectrum occupancy prediction. The proposed methods were simulated and the results show significant improvement in prediction error over local, and hard fusion-based spectrum prediction.