【摘要】：改革開放以來我國(guó)銀行業(yè)迅速發(fā)展，個(gè)人信用評(píng)估在銀行業(yè)的重要性日顯突出。目前國(guó)內(nèi)外的個(gè)人信用評(píng)分模型以統(tǒng)計(jì)學(xué)模型和人工智能方法為主。這些方法各有利弊，統(tǒng)計(jì)學(xué)方法可以提供假設(shè)檢驗(yàn)但精度不高，人工智能方法精度較高但解釋性不強(qiáng)。而且，這些成熟個(gè)人信用評(píng)分模型都面臨著拒絕推論和信用樣本動(dòng)態(tài)變化的問題。目前大多評(píng)分機(jī)構(gòu)僅用已獲得貸款的部分客戶作為樣本訓(xùn)練模型來預(yù)測(cè)整體信用客戶人群，這將導(dǎo)致非隨機(jī)性樣本偏差，直接影響評(píng)分模型的有效性；拒絕推論就是對(duì)這種樣本偏差的糾正。信用樣本動(dòng)態(tài)變化指信用樣本由于各種因素造成個(gè)人信用狀態(tài)發(fā)生變化或信用人群由于經(jīng)濟(jì)社會(huì)的發(fā)展而發(fā)生整體漂移，，這會(huì)使得信用評(píng)價(jià)模型的結(jié)果和現(xiàn)實(shí)出現(xiàn)越來越大的偏差。拒絕推論和信用樣本動(dòng)態(tài)變化是個(gè)人信用評(píng)分領(lǐng)域中亟待解決的問題。 案例推理模擬人類大腦認(rèn)知過程，具有很強(qiáng)的理論基礎(chǔ)和廣泛的應(yīng)用背景，有望成為能夠解決拒絕推論的動(dòng)態(tài)信用評(píng)分模型。首先，本文根據(jù)案例推理的發(fā)展現(xiàn)狀，構(gòu)建了傳統(tǒng)案例推理信用評(píng)分系統(tǒng)。通過該系統(tǒng)的應(yīng)用發(fā)現(xiàn)案例推理在我國(guó)個(gè)人信用評(píng)估中既具有優(yōu)越性同時(shí)也存在著局限性，這些局限性包括現(xiàn)有銀行數(shù)據(jù)的影響和傳統(tǒng)案例推理假設(shè)的制約。其次，針對(duì)這些現(xiàn)實(shí)制約因素，分別從案例庫和案例推理循環(huán)兩個(gè)層面對(duì)案例推理系統(tǒng)進(jìn)行了優(yōu)化。案例庫的優(yōu)化包括案例表達(dá)的優(yōu)化、拒絕樣本的引入和案例庫的動(dòng)態(tài)優(yōu)化；推理循環(huán)的優(yōu)化包括了神經(jīng)網(wǎng)絡(luò)與K最近鄰法的混合案例檢索方法和貝葉斯案例重用方法。最后，利用深圳某商業(yè)銀行的個(gè)人信用數(shù)據(jù)對(duì)優(yōu)化的案例推理系統(tǒng)進(jìn)行了應(yīng)用。結(jié)果表明，優(yōu)化后的案例推理系統(tǒng)對(duì)拒絕樣本的識(shí)別能力顯著增強(qiáng)，能夠很好地處理拒絕推論和信用動(dòng)態(tài)變化問題；優(yōu)化的案例推理系統(tǒng)較傳統(tǒng)案例推理系統(tǒng)在準(zhǔn)確性上有所提高，而且在穩(wěn)定性和解釋性上有了很大改進(jìn)；優(yōu)化的案例推理系統(tǒng)是一個(gè)能夠和商業(yè)銀行信用政策相互動(dòng)態(tài)適應(yīng)的個(gè)人信用評(píng)分方法，對(duì)銀行的信貸政策具有政策支持性。
[Abstract]:Since the reform and opening up, China's banking industry has developed rapidly, and the importance of personal credit evaluation in the banking industry has become increasingly prominent. At present, the personal credit scoring models at home and abroad are mainly statistical models and artificial intelligence methods. These methods have their own advantages and disadvantages. Statistical methods can provide hypothetical test but the accuracy is not high, artificial intelligence method has higher precision but not strong explanation. Moreover, these mature personal credit scoring models are faced with the problems of rejection inference and dynamic changes of credit samples. At present, most rating organizations only use some clients who have received loans as sample training model to predict the whole credit customer population, which will lead to non-random sample deviation, which directly affects the effectiveness of the rating model. Rejection of inference is the correction of this sample bias. The dynamic change of credit sample refers to the change of individual credit status caused by various factors or the overall drift of credit population due to the development of economy and society. This will make the results of credit evaluation model and reality appear more and more big deviation. Rejection inference and dynamic change of credit samples are the problems to be solved in the field of personal credit scoring. Case-based reasoning (CBR), which simulates the cognitive process of human brain, has strong theoretical foundation and extensive application background, and it is expected to be a dynamic credit scoring model which can solve the problem of rejection inference. Firstly, according to the development of CBR, this paper constructs the traditional CBR credit scoring system. Through the application of this system, it is found that Case-Based reasoning (CBR) has both advantages and limitations in personal credit assessment in China. These limitations include the influence of existing bank data and the restriction of traditional Case-Based reasoning hypothesis. Secondly, the case-based reasoning system is optimized from two aspects: case base and case-based reasoning cycle. The optimization of case base includes the optimization of case representation, the introduction of rejected samples and the dynamic optimization of case base. The optimization of reasoning cycle includes the hybrid case retrieval method of neural network and K-nearest neighbor method and Bayesian case reuse method. Finally, the optimized case-based reasoning system is applied using the personal credit data of a commercial bank in Shenzhen. The results show that the optimized Case-Based reasoning (CBR) system can effectively deal with the problem of rejection inference and the dynamic change of credit. The optimized CBR system is more accurate than the traditional CBR system, and has great improvement in stability and explanation. The optimized Case-Based reasoning system (CBR) is a kind of personal credit scoring method which can dynamically adapt to the credit policy of commercial banks and has policy support to the credit policies of banks.
【學(xué)位授予單位】：哈爾濱工業(yè)大學(xué)
【學(xué)位級(jí)別】：碩士
【學(xué)位授予年份】：2012
【分類號(hào)】：F832.479;F224

【參考文獻(xiàn)】

相關(guān)期刊論文 前10條

1 陳為民;馬超群;馬林;;我國(guó)個(gè)人信用評(píng)分的發(fā)展趨勢(shì)[J];商業(yè)研究;2010年01期

2 杜淼淼;;美國(guó)個(gè)人信用評(píng)分系統(tǒng)及其啟示[J];南方金融;2008年08期

3 鄧超;胡威;唐瑩;;基于拒絕推論的小企業(yè)信用評(píng)分模型研究[J];國(guó)際金融研究;2011年04期

4 莊傳禮;;我國(guó)信用局個(gè)人信用評(píng)分發(fā)展研究[J];征信;2011年01期

5 殷爽;姜明輝;;基于DEA的個(gè)人信用評(píng)估方法[J];經(jīng)濟(jì)研究導(dǎo)刊;2008年12期

6 房文娟,李紹穩(wěn),袁媛,汪偉偉;基于案例推理技術(shù)的研究與應(yīng)用[J];農(nóng)業(yè)網(wǎng)絡(luò)信息;2005年01期

7 吳麗娟;張健宇;高立新;;基于神經(jīng)網(wǎng)絡(luò)和案例推理的智能診斷系統(tǒng)綜述[J];機(jī)械設(shè)計(jì)與制造;2009年03期

8 洪遠(yuǎn)芳;鄒永福;;基于最近鄰法和支持向量機(jī)的個(gè)人信用評(píng)估方法[J];科技信息;2010年33期

9 柳玉;賁可榮;;案例推理的故障診斷技術(shù)研究綜述[J];計(jì)算機(jī)科學(xué)與探索;2011年10期

10 韓雪;馮玉強(qiáng);;基于案例推理的談判支持系統(tǒng)的研究[J];控制與決策;2008年07期

本文編號(hào)：2156588