.png){kind=icon}
.png){kind=icon}
Return
A multi-task test case optimization framework with integrated explainable AI for customer churn prediction
.png){kind=icon}
PRE
AI
DOI：10.1007/s11227-025-07816-4.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
This paper introduces a novel Test Case Optimization Framework for customer churn prediction, integrating multi-task deep learning and explainable artificial intelligence. The proposed model simultaneously predicts churn status, credit score category, and high-balance risk using a shared neural architecture with task-specific output heads. A systematic set of test cases was generated, each defined by distinct configurations of training epochs, batch size, loss weights, and focal loss strategies. The optimal model was selected based on an optimization score that combines F1-score, recall, and performance stability. Experiments conducted on a large-scale banking dataset of 1 million customers demonstrate the effectiveness of the framework. The model achieved 83.05% accuracy, 83.05% recall, and 83.37% F1-score on the churn task, outperforming classical baselines such as Logistic Regression, Support Vector Machine, and Multilayer Perceptrons. Additionally, it attained 96.20% and 98.75% accuracy on the auxiliary tasks of credit score classification and balance flag detection, respectively. To improve interpretability, a lightweight rule-based XAI module was embedded to generate global and local churn explanations, while a risk alert system proactively identified high-risk customers. The results highlight the framework’s potential for building accurate and trustworthy churn prediction systems in real-world applications. The computational demands of evaluating numerous deep multi-task configurations on a million customer dataset necessitate high-performance computing resources. By executing test case generation, training, and evaluation in parallel across multi-GPU or distributed HPC clusters, the framework reduces runtime from days to hours, enabling near real-time model updates. This is feasible because each test case configuration is independent, allowing concurrent execution of generation, training, and evaluation without any inter-process dependencies.
Keywords:
Customer churn prediction
Multi-task Learning
Test case optimization
Focal loss
Explainable AI
Deep neural network
Model selection
Churn risk alert
Rule-based explainability
Journal
T
IF:
0
Papers:
647
Citations:
0