應用粒化計算建構信用風險評估模型

Abstract

財務風險資料多具有類別不均的問題(class imbalance)。類別不均指的是資料中類別數量不對稱，分為多數類別(major class)與少數類別(minor class)。在此情況下若將全部資料皆納入訓練樣本進行建模，則可能發生多數類別準確率高，少數類別準確率過低的情況。目前雖然中外文獻提出多項風險評估模型，但多數模型仍採用減少多數抽樣法(under sampling)進行建模。此種方法的資料完整性不足，模型受抽樣樣本影響，可能產生模型失準的問題。本研究採用粒化計算的方法建立風險評估模型，建模時不需透過抽樣。除了應用粒化計算建構信用風險評估模型，本研究改善目前粒化計算過程的缺失，加入新指標，避免產生少數類別平均分散至多數類別資訊粒子，提升粒化計算風險評估模型之效率。最後與不同的抽樣法比較，可得粒化計算風險評估模型可有效降低類別不均資料分類準確率不對稱情形之結論，維持一定整體準確率下提升少數類別準確率，亦即提升風險評估模型中的違約借款客戶風險準確率。

Most of the finance risk data with a class imbalance problem. Class imbalanced data means the asymmetric categories of data, a data with class imbalance problem could be divided into two categories: major class data and minor class data. If we use all the imbalanced data without sampling, the accuracy of major class instances could be very well, but poor predictive ability to identify minority instances. Many risk assessment models have been developed in many studies, but most of them use sampling method to deal with the class imbalanced data. This study use “Granular Computing” model to tackle class imbalance problems. Using Granular computing to construct risk model could provide a better insight into the essence of data, and effectively solve class imbalance problems. In order to improve the lack of Granular Computing, and enhance the efficiency of credit risk modeling, this study adds a new index: “PM” to avoid a situation which minor class data spread to major class granular. In the end, the study compares the granular computing risk assessment model with several sampling methods. By calculation and compare of the accuracy, AUC and G-means, we can conclude that using granular computing credit assessment model would have same or even better result than sampling models.