基于树模型的财务报销审批预测  

Tree Model Based Prediction of Financial Reimbursement Approval

作　　者：刘春雨[1,2] 施卓敏 于建军[1] LIU Chunyu;SHI Zhuomin;YU Jianjun(Computer Network Information Center,Chinese Academy of Sciences,Beijing 100190,China;University of Chinese Academy of Sciences,Beijing 100049,China)

机构地区：[1]中国科学院计算机网络信息中心,北京100190 [2]中国科学院大学,北京100049

出　　处：《数据与计算发展前沿》2021年第2期60-67,共8页Frontiers of Data & Computing

摘　　要：【目的】针对目前科研院所财务报销不规范导致的反复审批等问题,本文研究通过预测结果提升报销审批效率。【方法】本文针对财务报销审批进行业务建模,形成可机器理解的报销审批脱敏数据,并根据实际业务特点构造变量特征与标签,采用随机森林对重构后的变量进行重要度分析。使用决策树、随机森林、梯度随机树以及XGBoost四种分类算法对报销审批结果进行预测。【结果】通过随机森林算法证实重构变量对于报销审批结果预测的可靠性。四种树模型根据重构后的训练数据集归纳出一组分类规则,采用该规则对未审批的报销单进行预测,通过预测结果从四种树模型中评定出最佳模型。【结论】文章基于树模型,通过构造随机森林辅助判断影响报销审批结果的关键因素,并选用树模型算法实现报销审批预测模型的构建,为树模型在报销审批预测中的应用提供了算法基础。[Objective]Nowadays,how to reduce repeated submissions caused by irregular reimbursements for financial reimbursement approval to improve the efficiency of financial reimbursement becomes a big issue in daily scientific research management of CAS institutes.This paper studies the use of prediction results to improve the efficiency of reimbursement approval.[Methods]This paper conducts a business model for financial reimbursement approval,obtains desensitized data for reimbursement approval that can be machine-understood,constructs variable characteristics and labels according to actual business characteristics,and then uses a random forest approach to analyze the importance of reconstructed variables.Decision tree,random forest,gradient random tree,and XGBoost algorithms are used to predict the reimbursement approval results.[Results]The importance analysis by constructing random forest makes the reconstruction variables more credible,provides reliable support for the approval results predicted by the subsequent four tree-model algorithms,and evaluates the best model from the results.[Conclusions]Based on the tree model,this paper identifies the key factors that affect the results of reimbursement approval and applies the machine learning algorithms to predict financial reimbursement approvals,which provides an application basis for tree models in predicting reimbursement approval.

关 键 词：变量重构 数据挖掘 树模型 审批预测 

分 类 号：TP181[自动化与计算机技术—控制理论与控制工程]