(3.236.214.19) 您好！臺灣時間：2021/05/10 07:58

詳目顯示:::

:

• 被引用:1
• 點閱:148
• 評分:
• 下載:0
• 書目收藏:2
 For finance institutes, credit scoring has become the important issue that banks used to assess whether customers may pass due delinquency or not. With the development of financial market liberalization and Internet services to flourish, we are in the financially big data environment. How to use scientific methods to handle and analyze large amount of data have become a new issue faced by the banks. The study is to explore how to apply a modified logistic regression to solve the credit scoring problems. With the logistic regression method, we combine it with the stochastic gradient descent algorithm to reach the target function optimization. The consolidation method can help banks minimize the customers’ credit risks in a huge amount of data and construct an objective credit scoring model. In addition, the study also compared the logistic regression analysis in order to investigate the credit scoring models which were established by the preferred classification method. In the Hadoop cloud computing environment, we show that the application of modified logistic regression algorithm can effectively upgrade classification accuracy. Whether in the original attributes or the filter attributes, the proposed algorithm outperforms logistic regression. Both of them get accurate rate of 86% by credit scoring prediction models. Simultaneously, the modified logistic regression models are effective in reducing Type I and Type II errors. They have the lower cost in modeling time.
 中文摘要i英文摘要ii致謝iii目錄iv表目錄vi圖目錄vii第一章 緒論11.1研究背景與動機11.2研究目的21.3研究流程3第二章 文獻探討52.1信用評分(Credit Scoring)52.2主成分分析(Principal Component Analysis, PCA)62.3隨機梯度下降(Stochastic Gradient Descent, SGD)82.4邏輯斯迴歸(Logistic Regression)102.4.1多元邏輯斯迴歸14第三章 研究方法153.1分析模型及研究架構153.2資料預處理(Data Preprocessing)173.2.1平衡資料173.3主成分萃取193.4邏輯斯迴歸分類203.4.1最大概似估計(Maximum Likelihood Estimate)203.4.2先驗估計(Prior Estimate)213.4.3誤差函數(Error Function)233.5邏輯斯迴歸與SGD演算法243.6傳統的邏輯斯迴歸運作模式263.7信用評分模型之準確率衡量283.7.1預測模型評估 283.7.2分類準確率衡量303.7.3AUC評估準則30第四章 實驗結果與數據分析324.1實驗環境324.2資料來源344.2.1屬性變數描述344.3實驗結果產生過程364.4實證分析394.4.1原始屬性之實驗結果394.4.2篩選屬性後之實驗結果414.4.3分析結果比較444.4.4信用評分預測模型48第五章 結論495.1研究結論495.2管理意涵505.3研究限制與未來建議50參考文獻51附錄56符號彙編58