نوع مقاله : مقاله پژوهشی
نویسندگان
1 گروه مدیریت بازرگانی، دانشکده مدیریت دانشگاه خوارزمی، تهران، ایران
2 دانشجوی دکتری مدیریت بازاریابی، دانشکده مدیریت دانشگاه خوارزمی، تهران، ایران
چکیده
کلیدواژهها
موضوعات
عنوان مقاله [English]
نویسندگان [English]
In today's competitive landscape across various industries, retaining existing customers is far less costly and more profitable than acquiring new ones, making timely prediction of customer churn a strategic necessity. This study aims to present and evaluate an innovative incremental learning framework for predicting customer churn and to conduct a systematic comparison between Logistic Regression and a Multi-Layer Perceptron (MLP) neural network. To this end, real-world historical data from an automobile dealership in Mashhad was utilized, with a sample size of 47,082 transactions belonging to 3,684 unique customers, while ensuring brand confidentiality. After merging and engineering demographic, transactional, and behavioral features, an incremental learning framework with an expanding window strategy was employed across five stages to simulate the operational environment and prevent temporal data leakage.
The numerical findings indicate that the Logistic Regression model, on average across all stages, demonstrated numerically better performance compared to the MLP model on this specific dataset: an F1-score of 0.834 versus 0.568; and an ROC-AUC of 0.9547 versus 0.6286. This research demonstrates that in the dynamic environment of the automotive industry, applying a time-aware incremental learning framework alongside a simple and interpretable model such as Logistic Regression, provides a practical and effective solution for timely customer churn prediction. Furthermore, it offers a data-driven foundation for designing targeted customer retention interventions and optimizing the allocation of marketing and after-sales service resources for automotive industry managers.
کلیدواژهها [English]
)