【作者】 张硕；

【导师】 付永强；

【作者基本信息】 哈尔滨工业大学 ， 应用统计（专业学位）， 2025， 硕士

【摘要】 本研究聚焦上海市住房租赁价格,旨在剖析其影响因素并构建精准预测模型,为住房租赁市场各主体提供决策依据,推动市场健康发展。在阐述房地产价格影响因素理论基础上,对比分析多元线性回归、随机森林、XGBoost极端梯度提升和LightGBM轻量梯度提升模型在住房租赁价格预测中的应用。研究数据源于贝壳网站2024年10月上海市12000条租房数据及百度地图POI信息。经数据清洗、变换等预处理,有效降低数据维度与复杂度,使数据更契合模型分析需求。采用独热编码处理分类型数据,运用对数变换消除量纲影响并改善数据分布。多元线性回归模型分析显示,精装、近地铁等部分特征变量显著影响租金,决定系数R~2为0.8482,能解释84.82%的目标变量变异,平均绝对百分比误差MAPE为1.6232%。随机森林模型中,面积、地铁站数量等是关键影响因素,其决定系数R~2达0.8811,MAPE为1.5263%,在租金中段预测精准。XGBoost极端梯度提升模型经参数调优,决定系数R~2为0.9022,MAPE为1.4378%,预测效果良好,租赁方式、面积等因素对租金影响突出。LightGBM轻量梯度提升模型损失梯度下降速度快,决定系数R~2为0.8829,MAPE为1.5743%,特征重要性排序表明人们租房时优先考虑房源周边综合条件。综合对比,XGBoost极端梯度提升模型在解释目标变量变异方面表现最佳,能更精准捕捉租金影响因素及规律,为各市场主体提供更可靠的决策参考。本研究不仅丰富了住房租赁领域理论,还为政府制定政策、企业优化运营、租客房东合理决策提供有力支持,对促进住房租赁市场规范化、透明化和可持续发展意义重大。更多还原

【Abstract】 This study centers on the housing rental prices in Shanghai,with the aim of dissecting their influencing factors and constructing precise prediction models.The findings are intended to offer decision-making guidance for all stakeholders in the housing rental market,thereby facilitating the market’s healthy development.By expounding on the theoretical underpinnings of real estate price-influencing factors,this research conducts a comparative analysis of the applications of multiple linear regression,random forest,XGBoost,and LightGBM models in predicting housing rental prices.The research data is sourced from 12,000 rental housing listings in Shanghai on the Ke.com website in October 2024,as well as POI information from Baidu Maps.Through data pre-processing steps such as data cleaning and transformation,the dimensionality and complexity of the data are effectively reduced,rendering it more suitable for model-based analysis.One-Hot Encoding is employed to handle categorical data,and logarithmic transformation is utilized to mitigate the impact of dimensionality and enhance the data distribution.The analysis of the multiple linear regression model reveals that certain characteristic variables,such as being finely decorated,being close to the subway,etc.,have a significant influence on rental prices.The coefficient of determination,R~2,stands at 0.8482,indicating that the model can explain 84.82%of the variance in the target variable.The mean absolute percentage error,MAPE,is 1.6232%.In the random forest model,factors like the property area and the number of subway stations in the vicinity are key determinants.Its R~2value reaches 0.8811,and the MAPE is 1.5263%.This model demonstrates high accuracy in predicting mid-range rental prices.After parameter tuning,the XGBoost model exhibits an R~2of 0.9022 and a MAPE of 1.4378%,yielding excellent prediction results.Factors such as the rental method and property area play a crucial role in determining rental prices.The LightGBM model features a rapid loss gradient descent rate,with an R~2of 0.8829 and a MAPE of 1.5743%.The importance ranking of features indicates that when people are looking for rental housing,they tend to prioritize the comprehensive conditions of the surrounding area.Through a comprehensive comparison,it is evident that the XGBoost model outperforms others in explaining the variance of the target variable.It can more accurately identify the influencing factors and underlying patterns of rental prices,thus providing more reliable decision-making references for all market participants.This research not only enriches the theoretical framework in the field of housing rental but also offers substantial support for the government in policy-making,enterprises in operational optimization,and renters and landlords in making informed decisions.It is of great significance for promoting the standardization,transparency,and sustainable development of the housing rental market.更多还原