摘要
社会救助服务是保障社会公平的重要举措之一。针对社会救助业务中被救助人员类型的精准识别问题具有极度不平衡性和传统算法在极度不平衡数据分类中具有强偏好性这两大难点进行研究,提出一种新的不平衡多分类模型—基于代价敏感的AdaBoost双层分类(Cost sensitive AdaBoost-Softmax,CA-SF)模型。首先,基于数量均衡原则设计一种“多变二”机制将多分类问题转化为二分类问题,利用具有代价敏感的AdaBoost模型以降低救助数据的极度不平衡性对分类效果的影响。其次,采用能有效规避多重共线性的Softmax回归来解决平衡数据的二次分类。综合实验结果表明CA-SF与多种常用模型相比在社会救助的精准识别上有更高的预测精度和更强的稳定性,能为精准社会救助提供科学且有效的辅助决策手段。
Social assistance service is one of the important measures to ensure social equity.Aiming at the two difficulties of accurate identification of the types of rescued persons in social assistance business,which are extremely unbalanced and the strong preference of traditional algorithms in extremely unbalanced data classification,a new unbalanced multi-classification model,Ada⁃Boost double-layer classification(Cost sensitive AdaBoost-Softmax,CA-SF)model based on cost sensitivity is proposed.Firstly,based on the principle of quantitative equilibrium,a"changeable two"mechanism was designed to transform the multi-classifica⁃tion problem into a binary classification problem,and the cost sensitive AdaBoost model is used to reduce the impact of the extreme imbalance of rescue data on the classification effect.Secondly,Softmax regression,which can effectively avoid multicollinearity,is used to solve the secondary classification of balanced data.The comprehensive experimental results show that CA-SF had higher pre⁃diction accuracy and stronger stability in the accurate identification of social assistance than a variety of commonly used models,and can provide scientific and effective auxiliary decision-making means for accurate social assistance.
作者
贺远珍
樊重俊
熊红林
HE Yuanzhen;FAN Chongjun;XIONG Honglin(School of Business,University of Shanghai for Science&Technology,Shanghai 200093;Antai College of Economics&Management,Shanghai Jiao Tong University,Shanghai 200240)
出处
《计算机与数字工程》
2023年第1期156-162,276,共8页
Computer & Digital Engineering
基金
教育部哲学社会科学研究重大课题攻关项目(编号:20JZD010)资助