New User Engagement Prediction Using Machine Learning Algorithms

This research paper presents our participation in the User Engagement Prediction Challenge, a competition hosted on the Zindi platform. The challenge's objective is to develop a predictive model that accurately predicts user activity in the subsequent month. Throughout the competition, we encountered the challenge of data imbalance, which we addressed by employing various techniques, including oversampling, undersampling, and ensemble learning. Among these approaches, ensemble learning proved to be the most effective in mitigating the data imbalance issue and improving prediction performance. In our study, we implemented different machine learning algorithms, including stack, hard voting, and CatBoost. We achieved a noteworthy F-score of 49.56% through extensive experimentation using the CatBoost algorithm. This achievement allowed us to secure the third position in the competition.