本研究旨在探討自建風控模型在降低成本和提高收益方面的應用。當前許多 公司依賴外部風控商進行風險評估，這導致了高成本和模型不透明等問題。本研究 提出了一種基於堆疊技術的自建風控模型，旨在利用內部數據建立準確且高效的 風控評分卡模型，以取代外部供應商並提高整體收益。 本論文的目標是提出一個風險控制模型，使用 Stacking 技術結合多種基底模 型（如邏輯迴歸、決策樹、XGBoost、LightGBM）達成目標並引入 LIME（Local Interpretable Model-agnostic Explanations）方法來提高模型解釋性。首先，收集公司 內部的貸款資料，並從中提取出用戶提交的相關信息，再利用模型輸出用戶違約機 率映射評分卡分數來調整貸款額度。 實驗結果顯示，自建風控模型在降低違約率和提升收益率方面表現優異，並且 相比外部風控模型有效降低了風控成本，提升了模型透明度和評估結果的精確性。 基於內部數據進行的風控模型在應對多變的市場需求和保障數據安全方面具有顯 著優勢。;This study aims to explore the application of self-built risk control models to reduce costs and increase revenue. Currently, many companies rely on external providers for risk assessment, leading to high costs and opaque models. This study proposes a self-built risk control model based on stacking technology, aiming to use internal data to establish an accurate and efficient risk scoring model to replace external providers and improve overall revenue. The goal of this thesis is to propose a risk control model that uses stacking technology combined with multiple base models (such as logistic regression, decision trees, XGBoost, and LightGBM) to achieve this goal. First, the company′s internal loan data is collected, and user-submitted loan information is extracted. Then, the model output probability is mapped to a scoring card, and the method is gradually adjusted and optimized. Experimental results show that the self-built risk control model performs excellently in reducing default rates and improving return rates. Compared to external risk control models, it effectively reduces risk control costs, improves model transparency, and enhances the accuracy of evaluation results. Risk control models based on internal data have significant advantages in responding to changing market demands and ensuring data security.
