计算机科学 ›› 2024, Vol. 51 ›› Issue (12): 259-268.doi: 10.11896/jsjkx.240300047
王宝财, 吴国伟
WANG Baocai, WU Guowei
摘要： 机器学习技术在金融领域的应用越来越多,为用户提供可解释的机器学习方法已成为一个重要的研究课题。近年来,反事实解释引起了广泛关注,它通过提供扰动向量来改变分类器得到的预测结果,从而提高机器学习模型的可解释性。但现有方法存在生成的反事实用例缺乏可行性和可操作性的问题。文中提出了一种新的反事实解释框架,通过引入特征变量代价权重矩阵的概念,考虑不同特征变量改变的难易程度,使得反事实结果更符合实际情况并更具可行性。同时,通过专家预定义特征变量代价权重矩阵的方式,提出了一种计算特征变量代价权重的可行方法,并允许用户根据实际情况进行个性化调整。定义的目标函数综合考虑了特征加权距离、稀疏性和接近性3个指标,确保了反事实结果的可行性、简洁性和接近原始样本集的性质。采用遗传算法来求解问题,进而生成最佳的行动方案。通过对真实数据集进行实验,证实了所提方法相比现有的反事实方法能够生成可行性和可操作性更强的反事实用例。
中图分类号:
| [1]ALA’RAJ M,ABBOD M.Classifiers consensus system ap-proach for credit scoring[J].Knowledge-Based Systems,2016,104:89-105. [2]ZHANG M Y.Research on Credit Risk Management in Banking Under the New Situation[J].Chinese Journal of Business Ma-nagement,2016,10(14):15-16. [3]LEE T S,CHEN I F.A two-stage hybrid credit scoring modelusing artificial neural networks and multivariate adaptive regression splines[J].Expert Systems with Applications,2005,28(4):743-752. [4]KANG S,CHO S.Approximating support vector machine with artificial neural network for fast prediction[J].Expert Systems with Applications,2014,41(10):4989-4995. [5]MOHAMMADI N,ZANGENEH M.Customer credit risk as-sessment using artificial neural networks[J].International Journal of Information Technology and Computer Science,2016,8(3):58-66. [6]LIU X Y,QU Y W,ZHOU Q Y.Self-attention credit assessment model[J].Chinese Journal of Computer Engineering and Applications,2019,55(13):36-41. [7]YU L,WANG S Y,LAI K K.Credit risk assessment with a multistage neural network ensemble learning approach[J].Expert Systems with Applications,2008,34(2):1434-1444. [8]ZHOU M X.Study on User Profiling Based on Deep NeuralNetworks[D].Changsha:Hunan University,2018. [9]MELIS D A,JAAKKOLA T.Towards robust interpretabilitywith self-explaining neural networks[C]//Proceedings of the 32nd Int Conf on Neural Information Processing Systems.USA:Curran Associates Inc.,2018:7775-7784. [10]POULIN B,EISNER R,SZAFRON D,et al.Visual explanation of evidence with additive classifiers[C]//Proceedings of the 18th Conf on Innovative Applications of Artificial Intelligence.Palo Alto,CA:AAAI Press,2006:1822-1829. [11]KONONENKO I.An efficient explanation of individual classifications using game theory[J].Journal of Machine Learning Research,2010,11(Jan):1-18. [12]HAUFE S,MEINECKE F,GÖRGEN K,et al.On the interpretation of weight vectors of linear models in multivariate neuroimaging[J].NeuroImage,2014,87:96-110. [13]HUYSMANS J,DEJAEGER K,MUES C,et al.An empiricalevaluation of the comprehensibility of decision table,tree and rule based predictive models[J].Decision Support Systems,2011,51(1):141-154. [14]BRESLOW L A,AHA D W.Simplifying decision trees:A survey[J].The Knowledge Engineering Review,1997,12(1):1-40. [15]KONG X W,YANG H.A defense method against adversarialexamples based on the interpretability of deep neural network models:China,CN112364885A[P].2021-02-12. [16]RIBEIRO M T,SINGH S,GUESTRIN C.“Why should I trust you?” Explaining the predictions of any classifier[C]//Procee-dings of the 22nd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining.USA:ACM Press,2016:1135-1144. [17]ZHOU Z H,JIANG Y,CHEN S F.Extracting symbolic rulesfrom trained neural network ensembles[J].AI Communications,2003,16(1):3-15. [18]LIN K,GAO Y.Model interpretability of financial fraud detection by group SHAP[J].Expert Systems with Applications,2022,210:118354. [19]RIBEIRO M T,SINGH S,GUESTRIN C.Anchors:High-precision model-agnostic explanations[C]//Proceedings of the 32nd AAAI Conf on Artificial Intelligence.Palo Alto,CA:AAAI Press,2018. [20]PEARL J.Theoretical impediments to machine learning withseven sparks from the causal revolution[J].arXiv:1801.04016,2018. [21]RODRIGUEZ P,CACCIA M,LACOSTE A,et al.Beyond Tri-vial Counterfactual Explanations with Diverse Valuable Explanations[C]//Proceedings of the International Conference on Computer Vision(ICCV).2022:1036-1045. [22]DEL SER J,BARREDO-ARRIETA,DÍAZ-RODRÍGUEZ N,et al.On generating trustworthy counterfactual explanations[J].Information Sciences,2024,655:119898. [23]POYIADZI R,SOKOL K,SANTOS-RODRIGUEZ,et al.FACE:Feasible and Actionable Counterfactual Explanations[C]//Proceedings of the AAAI/ACM Conference on AI,Ethics,and Society(AIES).New York:ACM,2020. [24]KANAMORI K,TAKAGI T,KOBAYASHI,et al.DACE:Distribution-aware counterfactual explanation by mixed-integer li-near optimization[C]//Proceedings of the IJCAI International Joint Conference on Artificial Intelligence.2020:2855-2862. [25]BERK U,ALEXANDER S,YANG L.Actionable recourse inlinear classification[C]//Proceedings of the Conference on Fairness,Accountability,and Transparency.2019:10-19. [26]CUI Z C,CHEN W L,HE Y J,et al.Optimal action extraction for random forests and boosted trees[C]//Proceedings of the 21th ACM SIGKDD International Conference on Knowledge Discovery and Data Mining.USA:ACM,2015:179-188. [27]VERMA S,DICKERSON J,HINES K.A Review of Counter-factual Explanations for Machine Learning[J].arXiv:2020:1-13. [28]BREUNIG M M,KRIEGEL H,NG R T,et al.LOF:Identifying Density-Based Local Outliers[C]//Proceedings of the 2000 ACM SIGMOD International Conference on Management of Data.USA:ACM,2000:4-23. [29]GUIDOTTI R,MONREALE A,RUGGIERI S,et al.Factualand Counterfactual Explanations for Black Box Decision Making[J].IEEE Intelligent Systems,2019,34(6):14-23. [30]DANDL S,MOLNAR C,BINDER M,et al.Multi-ObjectiveCounterfactual Explanations[C]//Proceedings of the International Conference on Parallel Problem Solving from Nature.2020:448-469. [31]WACHTER S,MITTELSTADT B,RUSSELL C.Counterfa-ctual explanations without opening the black box:Automated decisions and the GDPR[J].Harvard Journal of Law & Technology,2018,31(2):842-887. |
|
||
首页