A novel financial performance metric to minimize misclassification costs in model selection
提出一种新型财务绩效指标（FPM），用于最小化信用风险评估中假阳性和假阴性导致的误分类成本，在德国信用数据集上验证其比传统指标更准确，可为机构带来显著财务收益。
Abstract A novel financial performance metric (FPM) is introduced seeking to minimise the misclassification cost arising from false positives and false negatives in credit risk assessment. Using the German Credit Dataset (GCD), important financial variables are simulated according to four different asset classes to enable a more accurate and reliable, multidimensional model selection. The misclassification cost arising from FPM is compared with commonly used statistical metrics and the credit scoring example dependent cost matrix (CSEDCM) metric. The results show that CSEDCM underestimates false prediction costs by as much as 99% compared to the FPM. A range of high- performance machine learning methods was compared using FPM and statistical metrics. The Multi-Layer Perceptron outperformed other methods on statistical metrics and overall on financial costs, while a mix of algorithms worked best on either side of the decision threshold. The results confirmed that the proposed FPM would provide a significant financial benefit to organisations.