Predicting user conversion from free trial to paid subscription is an important task for video streaming platforms. It helps to improve marketing efforts and reduce unnecessary costs. Traditional marketing methods, such as sending many emails, often give low results and can cause users to lose interest. This study is based on behavioral data from 60,000 trial users of a video streaming platform, of whom about 10% converted to paid subscriptions. The dataset includes user metadata, session activity, and conversion outcomes, providing rich signals of viewing behavior. In this paper, we present a machine learning approach to predict the chance of conversion by analyzing user behavior during the trial period. We test both single-model and ensemble-model classifiers, using features based on session activity, device usage, and content variety. Since only 10% of users convert, we tested different sampling strategies, including SMOTE, undersampling on 10,000 samples, and a combined SMOTE + undersampling approach. The results show that ensemble methods, especially Random Forest, perform better than simpler models in terms of F1-score and recall for the minority class. Furthermore, the feature importance analysis shows that session duration, number of used devices, and diversity of content are strong indicators of conversion. These findings show that machine learning can support more effective and targeted marketing, with less unwanted communication.