Technical efficiency-based selection of learning cases to improve forecasting of neural networks under monotonicity assumption