Corporate Credit Rating Prediction System

Project Description

This project aims to predict the Corporate Credit Rating, which represents an independent agency's opinion on a company's ability to meet its financial obligations. Using supervised learning and probabilistic reasoning techniques, the system automates corporate risk assessment, classifying companies based on their financial soundness.

The project uses Machine Learning models such as Decision Tree, RandomForest, XGBoost, and LightGBM to predict the rating, and includes a Bayesian network to make probabilistic inferences on the data.

Project Structure

1. Data Preprocessing

   • Cleaning and preparing the dataset using techniques like normalization and one-hot encoding.
   • Reducing the rating classes from 22 to 4 main categories representing minimum-low, medium, medium-high, and high-default risk.

2. Supervised Learning

   • Use of models such as Decision Tree, RandomForest, XGBoost, and LightGBM.
   • Testing various class rebalancing techniques, such as SMOTE and ADASYN.

3. Bayesian Network

   • Creation of a Bayesian network to handle cases of missing data and generate new predictions based on probabilistic evidence.

Results

The results show that the SMOTE-based approach yielded the best overall performance, particularly with the RandomForest and XGBoost models. The metrics used for evaluation include Balanced Accuracy, Cohen's Kappa, and Geometric Mean.

Future Developments

• Increase the amount of data to improve prediction accuracy.
• Further explore the use of neural networks for classification.
• Integrate additional variables for a more comprehensive and accurate analysis.