In the aim of developing business, companies will likely be done corporate actions. One of them is offering their shares to the public. A company lists its shares on the stock exchange and offers them to be traded in public for the first time through an initial public offering (IPO). In Indonesia, studies related to the IPO return prediction mainly focus on using the linear regression approach, which is sensitive to outlier data. In the last decades, machine learning has been widely introduced and proved to result in better performance in financial data cases. Recently, the applications of ensemble algorithms, which combine several machine learning algorithms, show better performances than single approaches. Therefore, this study aims to predict the performance of IPO by calculating the return using an ensemble learning approach. The ensemble methods employed are random forest and gradient boosted tree. IPO return predictions were conducted in two approaches, through short-term and long-term performance. In the short term, the initial return of IPO on the first offering day was predicted. For the long-term, a prediction was made to calculate the Buy and Hold Abnormal Return (BHAR) 36 months after the IPO. The results show that the predictive model of ensemble learning proved to have better performance than linear regression. However, there is no significant difference between the results of the ensemble bagging (random forest) and boosting (gradient boosted tree) models. © 2024 Author(s).