In-game purchase analytics is gaining importance in the light of personalized marketing strategies employed by gaming companies to enhance player engagement. In that regard, the present study identifies player activity patterns incepting purchases within the game using machine learning approach. Specifically, Extra Trees algorithm is evaluated for its performance in in-game purchase prediction when compared with another ensemble model, AdaBoost classifier. Extra Trees, a decision tree-based ensemble algorithm, outperforms AdaBoost with regard to its efficiency in evaluating high-dimensional data and robustness to overfitting, whereas the latter involves the use of weighted ensembles to minimize prediction errors. The experimental results demonstrate that the Extra Trees algorithm is superior to AdaBoost in terms of classification accuracy and speed. Extracting details of player interaction such as the ‘duration per game play’, ‘achievements’, and ‘prior purchases’ to define player activity helped Extra Trees to predict activity with more details and achieved a better predicting analysis than AdaBoost. The results also underline that the effectiveness of ensemble methods can be enhanced by analyzing the importance of various features using the Extra-Trees algorithm. Importantly, this study characterizes Extra-Trees as a reliable ensemble learning method in the context of gaming datasets with non-linear interactions or complex relations.