Predicting user churn—that is, when users stop participating—in non-subscription gig platforms such as food delivery or ride-hailing services, poses unique challenges due to the absence of explicit labels and the dynamic nature of user behavior. Existing methods often rely on aggregated behavioral snapshots or static visual representations, which obscure short-term temporal cues critical for early detection. In this work, we propose a temporally-aware computer vision framework that models user activity patterns as a sequence of radar chart images—circular plots that visually encode multivariate behavioral features at the daily level. By integrating a pretrained convolutional neural network (CNN) with a bidirectional LSTM, our architecture captures both spatial patterns within each day and temporal dynamics across time. Extensive experiments on a large real-world dataset demonstrate that our method outperforms classical models and Vision Transformer (ViT)-based radar chart baselines, yielding gains of +17.7 in F1 score, +29.4 in precision, and +16.1 in AUC. The framework’s modular design and efficient deployment characteristics make it well-suited for large-scale churn modeling in dynamic gig-economy settings.