Interpretable Cost-Sensitive Predictive Modeling for Churn-Oriented Decision Optimization in Business Analytics

This research presents an interpretable, cost-sensitive predictive modeling framework for early decision optimization in business analytics, with a specific focus on customer churn prediction. The study integrates traditional interpretable machine learning models, including Logistic Regression, Random Forest, and Gradient Boosting, with a business-oriented cost framework to move beyond conventional accuracy-based evaluation. Using a real-world customer dataset, the proposed framework evaluates models not only on standard performance metrics such as accuracy, precision, recall, and F1-score, but also on their financial impact measured through business loss. The results demonstrate that the model with the highest statistical accuracy is not necessarily the most optimal from a business decision perspective. Furthermore, the research emphasizes interpretability by analyzing feature-level contributions, enabling decision-makers to understand the drivers behind churn predictions and to design targeted intervention strategies. This work provides a practical, data-driven decision support approach that bridges the gap between predictive analytics and real-world business value, making it especially suitable for managerial and operational decision-making contexts.