As I understand it, the model is trained for classification and interpretation of time series data, but have you tried benchmarking it at forecasting? Explanation and recommendations are often deeply intertwined with forecasts, so there must be at least some effect there?