Title: ChurnNet: An Optimized Modern AI Framework for Churn Prediction

Abstract: Heightened market rivalry and the increasing homogeneity of retail offerings have significantly lowered the threshold for consumers to defect to competing brands. Consequently, precise churn forecasting has emerged as a critical asset for executing targeted, personalized marketing initiatives and mitigating customer attrition. This research assesses the efficacy of established machine learning algorithms—specifically Random Forests, XGBoost, and Support Vector Machines—and benchmarks them against the Unified Multi-Task Time Series Model, which addresses churn as a binary time-series classification problem. Although the latter model demonstrates a robust capability to capture intricate temporal patterns and inter-variable dependencies, our empirical results reveal that traditional approaches maintain superior predictive accuracy, data efficiency, and lower computational demands for both training and deployment in the context of churn prediction. These conclusions hold true across a variety of datasets and different methodologies for defining churn labels.

Source: arXiv Generated at: 2026-06-02 00:00:00 UTC