Cognitive Modelling of Bankruptcy Risk: A Comparative Analysis of Machine Learning Models to Predict the Bankruptcy

Machine learning models can assess the financial health of companies and predict the likelihood of them going bankrupt. Early detection gives companies and stakeholders more time to implement strategies to mitigate financial risks or take corrective actions to avoid bankruptcy. This can be particularly useful for companies as it helps them avoid potential financial difficulties, for example, the recent bankruptcy of Silicon Valley Bank (SVB) has led to market volatility, liquidity disruption, and economic instability. This paper compares machine learning models to determine which model predicts bankruptcy better. The dataset of 20 years of US company bankruptcy was obtained from Kaggle.com and consists of 78682 instances and 21 attributes. In this study, we applied robust preprocessing to increase the accuracy of bankruptcy prediction. It aids in determining significant factors contributing to operation uncertainty and helps regulators and investors forecast the probability of default for better risk management. We applied an 80:20 split for training and validation respectively in our dataset and followed proper tuning of parameters using cross-validation in the training set. We compared several performance matrices, including accuracy and ROC-AUC in various machine learning models such as logistic regression, KNN, decision tree, support vector machine, neural network, and random forest to demonstrate the validity of our study findings. The KNN Classifier has come up champion model with an accuracy of 94.41% and an ROC AUC of 80.45% among all machine learning models as better predictors for bankruptcy.