International Conference on Computer and Knowledge Engineering

Home / 15th International Conference on Computer and Knowledge Engineering

Uncertainty-Aware Deep Ensembles for Confident Customer Churn Prediction with Rejection Option

Authors :

Fatemeh Moradi¹ Mehran Tarif² Mohammadhossein Homaei³

1- Faculty of Engineering Isfahan (Khorasgan) Branch, Islamic Azad University Isfahan, Iran 2- Department of Computer Science University of Verona Verona, Italy 3- Media Engineering Group University of Extremadura C´aceres, Spain

Keywords :

Customer churn prediction،uncertainty quantification،deep ensembles،rejection option،Monte Carlo dropout،Bayesian neural networks

Abstract :

Customer churn prediction is a key challenge in business intelligence, especially in industries where retaining clients is costly and vital. Most prediction models achieve high accuracy but fail to express how confident their predictions are, which can lead to expensive or misguided interventions. To address this gap, this paper proposes an Uncertainty-Aware Ensemble (UA-Ensemble) framework that quantifies prediction confidence alongside churn prediction. It combines five distinct neural network architectures, including attention-based LSTMs, Bayesian networks, and Monte Carlo Dropout, to capture both aleatoric and epistemic uncertainty. The framework incorporates a cost-aware rejection mechanism to abstain from acting on unreliable predictions. Experiments on large-scale datasets from banking, e-commerce, and telecom sectors achieved 94.2% accuracy with well-calibrated uncertainty estimates and 18.3% reduction in intervention costs compared to baseline methods. The approach outperforms existing models including traditional machine learning methods, deep learning baselines, and alternative uncertainty quantification techniques, demonstrating effectiveness across different industries and data scales, making it a practical tool for trustworthy business decision-making.