Performance Evaluation of Machine Learning Models for Prediction and Analysis of Financial Distress

The importance of financial distress prediction (FDP) in addressing corporate financial risk management has grown over the last decade. The majority of earlier studies in this area attempted to address financial issues and bankruptcy forecasts through the use of AI technologies or conventional bankruptcy prediction models. Investors' decisions can be influenced by news of financial difficulty, and relying too much on the opinions of financial analysts or one's own subjective assessment can lead to poor decision-making. As a result, this study aims to provide a fresh, objective model of FDP. This study evaluate three ML models: LSTM, RF, and KNN, measuring their performance using critical metrics like F1Score, AUC, Precision, Accuracy, and Recall. A following models are implemented on the NYSE and NASDAQ from 1999 to 2018, utilizing a dataset comprising 8,262 firms. Our results indicate that the LSTM model significantly outperforms both RF and KNN, achieving an F1-score 77.2% and an AUC 83.2%. These findings suggest that LSTM is a superior approach for financial distress prediction, highlighting its potential for enhancing risk assessment frameworks in financial analytics.