关键词: 特征选择, 随机森林（RF）, 主成分分析（PCA）, 灰狼优化（GWO）算法, 模型堆叠

Abstract: The long-term customer stability is important for hotel revenue and increased competitiveness. In customer churn prediction research, data collected in the production environment has issues such as large volume, high dimensionality, and a lot of noise, resulting in decreased accuracy, stability, and generalization ability of machine learning models. Aiming at this type of problem, a feature selection method based on PCA and grey wolf optimizer (GWO) is designed, and a customer churn prediction model is established using model stacking technique. A method utilizing Pearson correlation coefficient and the feature importance of random forest to determine the feature subset for dimensionality reduction is proposed. In accordance with the characteristics of the integrated feature selection model, improvements are made to the gray wolf optimization algorithm (GWO) by enhancing the gray wolf position update mechanism and convergence criteria, which are employed in the process of selecting the optimal feature subset as specified in the paper. Ten different machine learning models are selected for training, and the model with the best F1-score performance is chosen as the base model. Meta-model training is conducted. Experimental results indicate that when using a certain hotel customer information dataset, the convergence speed of the improved GWO algorithm is significantly increased, and the F1-score of the prediction model reaches 97.9%. The model has strong generalization capabilities.

Key words: feature selection, random forest (RF), principal component analysis(PCA), grey wolf optimization (GWO) algorithm, model stacking