【摘要】：隨著國(guó)家經(jīng)濟(jì)的不斷發(fā)展,房地產(chǎn)業(yè)與人們的生活日益相關(guān)。不論是企業(yè)還是個(gè)人,都把投資房地產(chǎn)業(yè)作為資產(chǎn)投資的重要一項(xiàng)。至目前為止個(gè)人住房貸款已成為商業(yè)銀行的最大業(yè)務(wù)之一,因此企業(yè)和個(gè)人的還款能力成為商業(yè)銀行住房信貸風(fēng)險(xiǎn)管理中的一項(xiàng)重要參考指標(biāo)。 在國(guó)外,商業(yè)銀行在對(duì)客戶還款能力評(píng)估方面已經(jīng)進(jìn)行很長(zhǎng)時(shí)間的探索并取得很好的發(fā)展成果,目前商業(yè)銀行基本采用統(tǒng)計(jì)的方法對(duì)客戶還款能力進(jìn)行量化分析。在我國(guó),客戶信用評(píng)估系統(tǒng)的建立相對(duì)落后,并未形成一個(gè)完整的體系,因而銀行的信用風(fēng)險(xiǎn)較大。因此,加快商業(yè)銀行信用評(píng)估系統(tǒng)建設(shè)的任務(wù)迫在眉睫,從而為商業(yè)銀行提供科學(xué)的決策依據(jù)。 作為個(gè)人住房貸款還款信用評(píng)估的其中一種方法,神經(jīng)網(wǎng)絡(luò)模型以其自學(xué)習(xí)、自調(diào)整以及非線性映射的優(yōu)點(diǎn),被應(yīng)用于量化個(gè)人信用評(píng)估模型中,但是神經(jīng)網(wǎng)絡(luò)在個(gè)人信用住房貸款評(píng)估中的實(shí)際應(yīng)用并不是很廣泛,有待進(jìn)一步探索。在多種神經(jīng)網(wǎng)絡(luò)模型中,BP神經(jīng)網(wǎng)絡(luò)在神經(jīng)網(wǎng)絡(luò)中應(yīng)用較為廣泛,因?yàn)锽P神經(jīng)網(wǎng)絡(luò)的具有自適應(yīng)、較強(qiáng)泛化能力以及容錯(cuò)能力好的顯著優(yōu)點(diǎn)。本文利用某家商業(yè)銀行客戶住房貸款還款真實(shí)記錄信息,并選取每一項(xiàng)還款記錄中的11項(xiàng)評(píng)估指標(biāo)構(gòu)建評(píng)估體體系.然后在計(jì)算上利用MATLAB軟件對(duì)數(shù)據(jù)進(jìn)行仿真分析。同時(shí),隨著科研工作者對(duì)BP神經(jīng)網(wǎng)絡(luò)的研究不斷深入,至目前為止已經(jīng)產(chǎn)生了很多訓(xùn)練算法,對(duì)于這些訓(xùn)練算法性能優(yōu)劣評(píng)比應(yīng)該具體問(wèn)題具體分析,從而針對(duì)某一具體問(wèn)題找到其最最優(yōu)算法。本文在BP神經(jīng)網(wǎng)絡(luò)模型的設(shè)計(jì)過(guò)程中,分別使用附加動(dòng)量法、自適應(yīng)學(xué)習(xí)速率、擬牛頓法、共軛梯度法、LM算法對(duì)BP神經(jīng)網(wǎng)絡(luò)模型進(jìn)行訓(xùn)練,通過(guò)對(duì)比每個(gè)訓(xùn)練結(jié)果的誤差、訓(xùn)練次數(shù)、訓(xùn)練時(shí)間及收斂速度,進(jìn)而確定采用基于LM算法的BP神經(jīng)網(wǎng)絡(luò)。使用已經(jīng)訓(xùn)練好的最優(yōu)BP神經(jīng)網(wǎng)絡(luò)對(duì)重新選擇的新客戶的住房貸款信息進(jìn)行測(cè)試,然后比較期望值與預(yù)測(cè)值之間的誤差,考核該BP神經(jīng)網(wǎng)絡(luò)模型的泛化能力,試驗(yàn)結(jié)果表明該網(wǎng)絡(luò)模型泛化能力良好。在基于LM算法的BP神經(jīng)網(wǎng)絡(luò)基礎(chǔ)上,本文提出一種改進(jìn)初始權(quán)值和樣本數(shù)據(jù)隨機(jī)性選取的方法,通過(guò)誤判率的減小說(shuō)明基于LM算法的改進(jìn)BP神經(jīng)網(wǎng)絡(luò)的合理性及有效性,從而提供了個(gè)人住房信貸評(píng)價(jià)的可行解決方案。
[Abstract]:With the development of national economy, real estate industry is increasingly related to people's life. Whether enterprises or individuals, investment in real estate as an important asset investment. Up to now, personal housing loan has become one of the biggest business of commercial banks, so the repayment ability of enterprises and individuals has become an important reference index in housing credit risk management of commercial banks. In foreign countries, commercial banks have been exploring for a long time and have achieved good results in the evaluation of customer repayment ability. At present, commercial banks basically use statistical methods to quantify the repayment ability of customers. In our country, the establishment of customer credit evaluation system is relatively backward and does not form a complete system. Therefore, the task of speeding up the construction of credit evaluation system of commercial banks is urgent, thus providing scientific decision basis for commercial banks. As one of the methods to evaluate the repayment credit of personal housing loan, the neural network model is applied to the quantitative personal credit evaluation model because of its advantages of self-learning, self-adjustment and nonlinear mapping. However, the application of neural network in the evaluation of personal credit housing loan is not very extensive and needs further exploration. Among various neural network models, BP neural network is widely used in neural network, because BP neural network has the advantages of self-adaptation, strong generalization ability and good fault-tolerant ability. This paper uses the real record information of a commercial bank customer's housing loan repayment, and selects 11 evaluation indexes of each repayment record to construct the evaluation body system. Then MATLAB software is used to simulate and analyze the data. At the same time, with the further research of BP neural network, a lot of training algorithms have been produced so far. The evaluation of the performance of these training algorithms should be analyzed concretely. In order to find its optimal algorithm for a specific problem. In the course of designing the BP neural network model, we use the additional momentum method, adaptive learning rate, quasi-Newton method, conjugate gradient method and LM algorithm to train the BP neural network model, and compare the errors of each training result. The training times, training time and convergence rate are used to determine the BP neural network based on LM algorithm. Using the trained optimal BP neural network to test the housing loan information of the re-selected new customer, then compare the error between the expected value and the predicted value, and evaluate the generalization ability of the BP neural network model. The experimental results show that the generalization ability of the network model is good. On the basis of BP neural network based on LM algorithm, this paper proposes a method to improve the random selection of initial weight and sample data. The rationality and effectiveness of the improved BP neural network based on LM algorithm are illustrated by the reduction of misjudgment rate. Thus provides the individual housing credit appraisal feasible solution.
【學(xué)位授予單位】：中國(guó)科學(xué)技術(shù)大學(xué)
【學(xué)位級(jí)別】：碩士
【學(xué)位授予年份】：2014
【分類(lèi)號(hào)】：F832.45;TP183

【參考文獻(xiàn)】

相關(guān)期刊論文 前10條

1 李英爽;我國(guó)個(gè)人住房消費(fèi)信貸的發(fā)展現(xiàn)狀與思考[J];北京房地產(chǎn);2005年01期

2 鄭旭;;我國(guó)商業(yè)銀行信用風(fēng)險(xiǎn)識(shí)別的實(shí)證分析[J];中國(guó)城市經(jīng)濟(jì);2012年01期

3 馬懷娟;;我國(guó)個(gè)人住房信貸風(fēng)險(xiǎn)管理研究[J];河北金融;2010年11期

4 王春峰,李汶華;小樣本數(shù)據(jù)信用風(fēng)險(xiǎn)評(píng)估研究[J];管理科學(xué)學(xué)報(bào);2001年01期

5 詹永玖;付祥;;我國(guó)商業(yè)銀行個(gè)人信用評(píng)估問(wèn)題的思考[J];金融縱橫;2008年09期

6 李鐵鷹,崔艷;一種基于粗糙集理論的神經(jīng)網(wǎng)絡(luò)分類(lèi)器的設(shè)計(jì)[J];計(jì)算機(jī)工程與應(yīng)用;2005年32期

7 孔煜;魏鋒;張燕;;我國(guó)個(gè)人住房消費(fèi)信貸的影響因素分析[J];建筑經(jīng)濟(jì);2011年10期

8 王紅喜;;淺析我國(guó)商業(yè)銀行住房消費(fèi)信貸的發(fā)展[J];內(nèi)蒙古煤炭經(jīng)濟(jì);2006年05期

9 張道宏;張璇;尹成果;;基于BP神經(jīng)網(wǎng)絡(luò)的個(gè)人信用評(píng)估模型[J];情報(bào)雜志;2006年03期

10 胡望斌,朱東華,汪雪鋒;商業(yè)銀行個(gè)人信用風(fēng)險(xiǎn)等級(jí)評(píng)估與預(yù)測(cè)[J];商業(yè)時(shí)代;2005年09期

，

本文編號(hào)：2286702