摘要
可融资性是PPP可持续发展的前提和基础。针对PPP项目可融资性存在行业差异性高及评价类别型字段过多问题,为防止评估过程出现严重过拟合现象和因二次加工产生的误差,提出集成LightGBM-Blending算法。基于社会资本视角构建可融资性评估体系,借鉴集成思想按行业划分训练集,构建以LightGBM算法为基础的多基分类器,避免对类别特征进行二次处理,并通过网络爬虫爬取CPPPC项目数据进行模型训练及测试,设计Blending算法进行模型融合与优化。实验结果表明:相较于传统集成算法,集成LightGBM-Blending算法评估精确度更高,可有效解决高行业差异性等问题;Blending融合策略比其他赋权融合策略评估精度提升了5.76%;引入低样本行业和高特质化行业样本进行验证,测试结果中百分比误差为低、中区间样本占比约为79.5%和78.26%,该模型可有效应用于不同行业PPP可融资性评估,具有良好的泛化能力。
Financiability is a prerequisite and foundation for the sustainable development of PPP. In order to prevent serious overfitting and errors caused by secondary processing in the evaluation process, the ensemble LightGBM-Blending algorithm is proposed to address the problems of high industry variability and too many evaluation category-type fields in the financiability of PPP projects. Based on the social capital perspective, we construct a financiability evaluation system, divide the training set by industry with the integrated idea, and build a multi-base classifier based on the LightGBM algorithm to avoid secondary processing of category features;finally, we crawl project data in the CPPPC through a web crawler for model training and testing, and design the blending algorithm for model fusion and optimization. The experimental results show that: compared with the traditional ensemble algorithms, the ensemble LightGBM-Blending algorithm has higher evaluation accuracy and can effectively solve the problems such as high industry variability;the blending fusion strategy improves the evaluation accuracy about 5.76% compared with other empowerment fusion strategies;a low sample industry and a high idiosyncratic industry sample is introduced for validation, and the test results show that the proportion of the low error interval and the medium error interval samples is about 79.5% and 78.26%. The model has good generalization ability, it can be effectively used for PPP financiability evaluation in different field.
作者
沈俊鑫
吕佳历
程墙
张经阳
SHEN Junxin;LU Jiali;CHENG Qiang;ZHANG Jingyang(Faculty of Management and Economics,Kunming University of Science and Technology,Kunming 650093,China;Faculty of Economics and Management,BeiBu Gulf University,Qinzhou 535001,China)
出处
《中国软科学》
CSSCI
CSCD
北大核心
2022年第1期50-61,共12页
China Soft Science
基金
国家自然科学基金项目(71964018,72163018)
云南省省院省校合作项目(SYSX201911)
广西哲学社会科学规划研究项目(18BGL014)。
