Rare Event Classification Based on Binary Generalized Extreme Value-Additive Models

The minority class in imbalanced dataset classification can be seen as a rare event since its probability of occurring is extremely small. Extreme value theory can be used to solve the problem of extreme events classification, which produce the binary generalized extreme value regression model. By considering the more flexible effects of predictors, smoothing parameters can be used to create a robust model, which is called the generalized additive model. In this study, we compare the additive model as well as the use of extreme value theory to predict financial distress in industrial companies in Indonesia. The company is classified as financially distressed if the company has an Interest Coverage Ratio that is less than one or has a negative Return on Assets. The predictors used in modelling are activity, profitability, liquidity, and solvability ratio. The empirical result shows that in the additive model, the use of the extreme value approach is able to capture rare events in financial distress prediction. This additive model was able to increase the performance of the binary GEVR model. The best model to predict financial distress is the Binary Generalized Extreme Value Additive Model with an AUC of 0.9303. The significant variables in financial distress prediction are EBITA, ETD, and ROE. In conclusion, the binary generalized extreme value additive model was able to describe the characteristics of financial distress data and have better performance in making predictions.