Stacked Machine Learning and SHAP-Based Insights for Hotel Cancellation Prediction

Abstract

This research addresses the limitations of traditional hotel cancellation management systems that rely on historical trends, manual judgment, and rule-based assumptions. Such approaches lack predictive accuracy, are time-consuming, and fail to capture complex customer behavior patterns, often leading to revenue loss and inefficient inventory management. The primary objective of this work is to develop an automated and interpretable machine learning framework for accurate prediction of hotel booking cancellations and identification of key influencing factors. The proposed system evaluates multiple machine learning algorithms, including Logistic Regression, Decision Tree, Random Forest, Multi-Layer Perceptron (MLP), To enhance predictive performance, the best-performing models are combined using a stacking classifier with Logistic Regression, Logistic regression, Random forest, MLP as the meta-learner. Furthermore, Principal Component Analysis (PCA) is incorporated for feature selection and dimensionality reduction to improve computational efficiency. To ensure model transparency and support data-driven decision-making, SHAP (Shapley Additive explanations) is integrated to provide feature-level interpretability of cancellation predictions. Experimental results demonstrate that the proposed framework significantly improves prediction accuracy while offering actionable insights into customer behavior. The system enables hotel managers to make proactive, data-driven decisions for effective revenue optimization, inventory control, and targeted marketing strategies