A comparative study of feature selection and feature extraction methods for financial distress identification

Area of Science:

• Business & Economics
• Computer Science & Data Science

Background:

• Financial distress identification is vital for economic stability.
• Traditional methods relying solely on financial ratios are insufficient due to data expansion.
• Increased data dimensionality leads to sparse data and model overfitting.

Purpose of the Study:

• To propose an efficient financial distress classification framework using dimensionality reduction and machine learning.
• To identify an optimal subset of features that minimizes the loss function for financial distress prediction.
• To enhance the accuracy of financial distress classification models.

Main Methods:

• Compared 15 dimensionality reduction techniques and 17 machine learning models.
• Conducted 1,432 experiments on Lithuanian enterprise data (2015-2022).
• Utilized Random Forest mean decreasing Gini (RF_MDG) for feature selection.

Main Results:

• The artificial neural network (ANN) model achieved the highest Area Under the Curve (AUC) score.
• Optimal performance was obtained using 30 features selected by the RF_MDG technique.
• A novel feature extraction approach was introduced.

Conclusions:

• The proposed framework effectively reduces dimensionality and improves financial distress classification.
• Combining advanced feature selection and machine learning models enhances predictive accuracy.
• The study offers a robust methodology for identifying financial distress in enterprises.