stability-regularization

This paper introduces a training-time stability regularization penalty to improve forecast stability without sacrificing accuracy, evaluated on M5 retail demand data, showing improvements in Forecast Stability Score while maintaining RMSE within 0.72%.