粒化計算於信用放款風險評估之應用

Abstract

財務風險資料多具有類別不平衡的問題，傳統分類器面對此種資料類型，容易將所有資料歸類至多數類別之中，導致產生對所有資料的判別準確率高，但對少數類別的判別準確率卻非常低的模型。中外文獻已提出多種方法建構評估信用放款風險的模型，但大多數研究在建構模型時均以對資料抽樣或增生的方法改變兩類別資料的數量，使得模型受建模樣本的影響很大，甚至可能導致模型失準。本計畫提出以粒化計算的方法建構信用放款風險評估模型，在建模時不需因為抽樣而捨棄資料，而改以對資訊粒化的方式解決類別不平衡的問題。在粒化計算的方法方面，本計畫提出一個在粒化計算中用於衡量資訊粒化程度的新指標，以改進現行方法對資訊粒化程度的評估方式，進而提高分類器對於少數類別的辨識能力。本計畫並使用台灣某金融機構借貸實際案例進行驗證，由驗證結果可知，本研究所提出的方法確實具有良好的分類準確率，並可解決類別不對稱資料型態分類問題。本計畫將所建構的信用放款風險評估模型整理於EXCEL介面，利用VBA語言所撰寫之程式匯整資料，使得決策者可直接於Excel工作表中進行資料運算與整理，實用價值高。

Most financial risk data has the class imbalnce problem, which means one class has far more (or less) instances than the other. Traditional classifiers tend to produce high predictive accuracy over the majority class but poor accuracy over the minority class. Most studies used re-sampling strategy to balance the number of instances between classes when constructing models for assessing credit loan risk. However, the resulting model accuracy highly depends on the sample which sometimes leads to an un-effective model. In this study, Granular computing is used to construct a risk assessment model for credit loan to effectively solve the class imbalance problem by information granulation. A new index is introduced to determine a suitable level of granularity to enhance the ability of effectively identifying instance from the minority class. Data provided by a financial institute in Taiwan is used to demonstrate the effectiveness of the proposed approach. The computational results show that, comparing with sampling methods, the proposed method could construct a credit assessment model with good accuracy on both majority and minority class while still achieve high accuracy overall. The proposed model can be implemented via Excel worksheet using VBA program to organize data. It can be applied easily to assess loan risk in practice.