PREDICTING BANKRUPTCY WITH MACHINE LEARNING MODELS

PREDICTING BANKRUPTCY WITH MACHINE LEARNING MODELS

G Naga Sujini1, Runku Yasaswini 2

1 Department of CSE, Mahatma Gandhi Institute of Technology, Gandipet, Hyderabad, 500075, Telangana, India.

2 Department of IT, Mahatma Gandhi Institute of Technology, Gandipet, Hyderabad, 500075, Telangana, India.

Abstract - Bankruptcy prediction is a pivotal task in finance and accounting, offering stakeholders critical insights into a firm's solvency. This study explores multiple machine learning classifiers, including XGBoost, Random Forest, Decision Trees, Neural Networks, SVM, Naïve Bayes, and Logistic Regression, to predict corporate bankruptcy. Using the Polish bankruptcy dataset from UCI, various preprocessing techniques such as EM and K-NN imputation and SMOTE were applied to handle data imbalance and missing values. Extensive performance evaluations using cross-validation reveal that ensemble methods like XGBoost and Random Forest offer superior predictive power, providing a robust foundation for decision-making. The study also presents a web-based prediction tool developed using Flask for real-time assessment. Future research is directed towards integrating deep learning architectures and multi-source datasets for enhanced generalizability.

Key Words:  Bankruptcy Prediction, Machine Learning, Financial Distress, XGBoost, Random Forest, Data, SMOTE.