[Paper Review] A hybrid neural network model based an improved PSO and SA for bankruptcy prediction

Fatima Zahra Azayite, Saïd Achchab|arXiv (Cornell University)|Jan 31, 2019

Financial Distress and Bankruptcy Prediction18 references4 citations

TL;DR

This paper proposes a hybrid neural network model that integrates improved Particle Swarm Optimization (PSO) and Simulated Annealing (SA) to enhance bankruptcy prediction accuracy. By combining variable selection with an advanced training algorithm, the model effectively reduces local minima issues and improves convergence, especially under missing data conditions, outperforming baseline models in empirical evaluation.

ABSTRACT

Predicting firm's failure is one of the most interesting subjects for investors and decision makers. In this paper, a bankruptcy prediction model is proposed based on Artificial Neural networks (ANN). Taking into consideration that the choice of variables to discriminate between bankrupt and non-bankrupt firms influences significantly the model's accuracy and considering the problem of local minima, we propose a hybrid ANN based on variables selection techniques. Moreover, we evolve the convergence of Particle Swarm Optimization (PSO) by proposing a training algorithm based on an improved PSO and Simulated Annealing. A comparative performance study is reported, and the proposed hybrid model shows a high performance and convergence in the context of missing data.

Motivation & Objective

• To improve the accuracy of bankruptcy prediction models by integrating advanced optimization techniques with neural networks.
• To address the challenge of local minima in neural network training through an enhanced PSO and SA hybrid algorithm.
• To enhance model robustness in the presence of missing data by incorporating effective variable selection and optimization.
• To develop a hybrid framework that combines the strengths of PSO, SA, and ANN for superior predictive performance in financial distress detection.
• To validate the model’s effectiveness through comparative performance analysis against conventional approaches.

Proposed method

• The model employs a hybrid architecture combining Artificial Neural Networks (ANN) with variable selection techniques to improve discriminative power.
• An improved PSO algorithm is designed to accelerate convergence and avoid premature convergence to local minima.
• Simulated Annealing (SA) is integrated with PSO to enhance global search capability and escape local optima.
• The hybrid PSO-SA algorithm is used to train the ANN, optimizing network weights and improving generalization.
• Variable selection is applied prior to training to reduce dimensionality and enhance model interpretability and performance.
• The model is evaluated under missing data conditions to test robustness and convergence stability.

Experimental results

Research questions

• RQ1 Can a hybrid PSO-SA optimization algorithm improve the convergence and accuracy of neural networks in bankruptcy prediction?
• RQ2 How does the integration of variable selection techniques affect the performance of ANN-based bankruptcy prediction models?
• RQ3 To what extent does the proposed model maintain high performance under missing data conditions?
• RQ4 Does the improved PSO-SA algorithm effectively reduce the risk of local minima compared to standard PSO or SA alone?
• RQ5 How does the proposed hybrid model compare to traditional ANN and other optimization-based models in predicting firm bankruptcy?

Key findings

• The proposed hybrid model demonstrates superior convergence behavior compared to standard PSO and SA-based training methods.
• The integration of variable selection significantly enhances the model’s discriminative accuracy between bankrupt and non-bankrupt firms.
• The model maintains high predictive performance even when faced with missing data, indicating strong robustness.
• Empirical results show that the improved PSO-SA algorithm reduces the likelihood of converging to local minima, improving overall optimization efficiency.
• The hybrid model outperforms baseline models in terms of prediction accuracy and stability, as confirmed through comparative performance analysis.
• The model achieves high convergence speed and accuracy, confirming the effectiveness of combining PSO, SA, and ANN in financial distress prediction.