Abstract: In order to improve the investment decision making level of option investors, the Stacking ensemble learning framework was used to predict the rise and fall trend of the implied volatility of SSE 50ETF options. Four tree-based ensemble models, random forest (RF), AdaBoost, Gradient Boosting Decision Tree (GBDT) and XGBoost, were selected as the base classifier to train the original data and carry out embedded feature selection. Again, the feature selection results were introduced into the base classifier, and the Stacking ensemble model based on feature selection was constructed by changing the different input features of the base classifier. The experimental results show that the improved Stacking ensemble model achieves significantly improvement on the precision accuracy and F1-score compared with the base classifier and the traditional Stacking model, and achieves a more ideal prediction effect.