Cloud coverage is quite a challenging target. The final performance of the model might be a bit unsatisfying. Are there other features from the weather dataset that would be interesting/satisfying to predict?