机器学习
标杆管理
人工智能
加权
计算机科学
梯度升压
均方误差
特征选择
稳健性(进化)
预测建模
数据挖掘
支持向量机
人工神经网络
Boosting(机器学习)
范畴变量
梯度下降
公制(单位)
交叉验证
破产预测
水准点(测量)
集成学习
监督学习
熵(时间箭头)
托普西斯
可预测性
性能指标
绩效改进
预处理器
选型
马修斯相关系数
皮尔逊积矩相关系数
高斯过程
作者
Navid Mohammadi,Elham Razghandi
出处
期刊:Array
[Elsevier BV]
日期:2026-10-02
卷期号:32: 101293-101293
标识
DOI:10.1016/j.array.2026.101293
摘要
Predicting guest satisfaction in peer-to-peer accommodation is challenging given the multidimensional, subjective nature of online reviews. This paper proposes an integrative pattern recognition framework combining objective multi-criteria weighting with comparative machine learning to predict satisfaction patterns from Airbnb listings. Using 18,335 Istanbul listings from Inside Airbnb, the Correlation Coefficient and Standard Deviation (CCSD) method constructs a composite satisfaction index from six review dimensions, while the platform’s overall rating is reserved as an external benchmark to avoid circular target construction. Ten supervised learning algorithms—spanning linear, regularized, tree-based, ensemble, kernel, instance-based, and neural network families—are benchmarked against three reference baselines under a unified, leakage-free validation protocol. Since no single metric captures all deployment needs, algorithms are ranked via TOPSIS, balancing predictive accuracy, training efficiency, and interpretability. Ensemble methods achieve the highest accuracy on structured listing and host attributes, with Gradient Boosting performing best (R 2 = 0.173, RMSE = 0.556; 9.0% error reduction over the mean baseline), while regularized linear models follow closely at far lower computational cost. TOPSIS identifies Gradient Boosting as the preferred model, with rankings stable underweight perturbation and alternative-removal checks. The composite index shows strong convergent validity (Pearson r = 0.947 with the external benchmark) and robustness to alternative weighting schemes. External validation on 13,094 Athens listings reproduces both the index properties and the cross-family performance ordering. The study’s contribution is integrative: linking theory-informed measurement, cross-family predictive benchmarking, and transparent multi-criteria model selection within a single reproducible pipeline for platform-mediated hospitality analytics.
科研通智能强力驱动
Strongly Powered by AbleSci AI