On the relationship between oil and gas markets: a new forecasting framework based on a machine learning approach