Financial Distress Prediction: Principle Component Analysis and Artificial Neural Networks

Abstract

Financial distress prediction is vital in financial decision-making for industry practitioners, financial system users and policymakers. Most of the research work on bankruptcy prediction and credit scoring compare the prediction accuracy of a model on a specific dataset to other models applied on a different dataset. Those research work did not consider factors that affect the datasets. These factors include feature selection, the total number of instances, number of variables employed, training/testing ratio and they affect the accuracy of the model. Therefore, knowing these factors help in analyzing the models carefully before comparing the model's prediction accuracy. We proposed a hybrid model which combine principal component analysis (PCA) with neural network analysis ANN for bankruptcy prediction. The experimental set-up is conducted using a Polish companies bankruptcy dataset which is publicly online on the UCI database. The dataset was pass through different training and testing ratio. The experimental study shows that PCA-ANN with 32 principal components outperforms other models using cross-validation for the training and testing ratio. The aim is to confirm the model with the highest accuracy, and the best area under the ROC curve (AUC) base on the training/testing ratio. Hence, in this research work, we declared the best model based on the said dataset. We observed that the ROC curve (AUC) of PCA-ANN with 32 principal components convincingly outperformed other models in terms the training/testing ratio of 10 folds cross-validation.

Cite this work

No public copy is stored for this record, so only the metadata is cited.

Stored only when both papers are in the corpus — 15 citations exist upstream.

Papers that cite the same references — the relational fallback while embeddings are still backfilling.