Please use this identifier to cite or link to this item:
http://tdr.lib.ntu.edu.tw/jspui/handle/123456789/54139
Title: | P2P借貸之違約率預測:應用機器學習 Predicting the Default Rates for P2P Lending:An Application of Machine Learning |
Authors: | Che-Wei Kuo 郭哲瑋 |
Advisor: | 林建甫(Chien-Fu Lin) |
Keyword: | 機器學習,P2P借貸,金融科技,違約預測,特徵工程, Machine Learning,P2P Lending,Fintech,Default Rate Prediction,Feature Engineering, |
Publication Year : | 2020 |
Degree: | 碩士 |
Abstract: | P2P lending是一種私人間透過網路平台的資金借貸方式,任何自然人、法人或非法人組織,皆可透過P2P業者提供的資訊與媒合,提供或獲取資金並從中獲取經濟上的利益。在台灣,透過銀行取得貸款雖非難事,但仍有許多被銀行排除在外的可能,這種實際上的現狀,使得手續便捷又不受地域、時間限制的P2P平臺興起,也讓人對它的未來走向產生濃厚的興趣與關注。
本研究以羅吉斯迴歸、基本決策樹、隨機森林、梯度提升決策樹、自我適應增強法、極限梯度提升樹與Catboost Classifier共7種演算法來進行風險控制的最佳模型分析,並重複進行多次驗證使得結果可以有穩定的表現;經過比較後,以綜合表現上以Catboost Classifier的表現為最佳算法。 P2P lending is a way of borrowing and lending funds between private individuals through an online platform. Any natural person, legal person or unincorporated organization can provide or get funds and receive economic benefits from the intermediaries provided by P2P providers. Although it is not difficult to get loans through banks in Taiwan, some limitations still exist. Those limitations gave rise to P2P platforms which are convenient in procedures and not subject to geographic or time constraints. Therefore, the future of P2P lending markets is worth expecting and paying attention to. This study uses a total of 7 algorithms, including Logistic Regression, Basic Decision Tree, Random Forest, Gradient Boosting Decision Tree, Adaboost Classifier, XGB Classifier and Catboost Classifier to perform the best model for risk control, and improved by repeated verifications. After comparison, Catboost Classifier performed the best result. |
URI: | http://tdr.lib.ntu.edu.tw/jspui/handle/123456789/54139 |
DOI: | 10.6342/NTU202002327 |
Fulltext Rights: | 有償授權 |
Appears in Collections: | 經濟學系 |
Files in This Item:
File | Size | Format | |
---|---|---|---|
U0001-0308202022471200.pdf Restricted Access | 1.95 MB | Adobe PDF |
Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.