Optimization of Multi-Layer Perceptron in Ensemble Using Random Search for Bankruptcy Prediction

Abstract

A corporation is in financial trouble if it has money problems but has yet to be bankrupt. Companies' financial woes must be identified early to implement various bankruptcy avoidance strategies. This study discusses the optimization of the multi-layer perceptron ensemble stacking classifier hyperparameter by majority voting as a methodology for constructing a classification model for bankruptcy prediction (MLP-STM). The primary aim of this study is to create a reliable model for use in MLP-STM-based financial ratio analysis. The used MLP-STM model has successfully optimized the MLP hyperparameters, which is a substantial contribution and a novel aspect of our study. Training and evaluation data came from a multi-class dataset with labels such as "distress area," "grey area," "safe area," and "save area." The training dataset is pre-processed to be well accepted by the learning ensemble. All performance models are evaluated using their confusion metrics and the Area Under the Curve (AUC). The primary conclusion of this study is that a novel MLP-STM classification model with varying hyperparameters can effectively classify features for detecting the performance of financially troubled businesses. When compared to the MLP-BAM and MLP-BOM models, the MLP-STM model performed best, with a 97% accuracy rate and a 100% AUC value. The results of this research have important implications for the banking and finance industries, including developing an early warning system in the event of financial bankruptcy.