Early Warning System of Gain Ensemble Bagging on Multilabel Banking Bankruptcy Data

Abstract

This research aims to develop an early detection system to identify potential banking financial problems and prevent crises early. Financial growth is strongly influenced by a sound monetary system, with the banking industry as its foundation. Therefore, it is crucial to detect early symptoms of banking financial problems, ideally before a crisis occurs. The Altman method was applied to construct the dataset. This model is capable of classifying into three target classes: safe zone, gray zone, and bankrupt zone. Information gain is used to determine the most important financial characteristics. We combine an artificial neural network-based meta-learner with a Bagging Ensemble model optimized through random search for hyperparameters to improve performance. The implementation includes careful data preprocessing, critical indicator search, multi-label classification model construction, and comprehensive performance evaluation. The proposed system outperforms conventional methods and a single machine learning model in terms of accuracy improvement. The implications of this research are providing useful analytical support for stakeholders who want to develop an early detection system to identify potential banking financial problems and prevent crises early.