Improving Forecast of Binary Rare Events Data: A GAM-Based Approach