The SVM is employed to predict the profit function, where a rolling learning-prediction procedure is introduced to deal with the time variant property of the process.