Artificial metaplasticity neural network applied to credit scoring.

Area of Science:

• Computational intelligence
• Machine learning for finance
• Credit risk assessment

Background:

• Accurate credit default risk assessment is crucial for financial institutions.
• Existing Artificial Neural Network (ANN) models for credit scoring often yield high error rates.
• There is a need for improved ANN algorithms to enhance credit scoring accuracy.

Purpose of the Study:

• To introduce and evaluate a novel ANN training algorithm inspired by neuronal metaplasticity for credit scoring.
• To address the challenge of high error rates in current credit scoring models.
• To demonstrate the efficacy of the proposed algorithm in handling imbalanced datasets and low-probability events.

Main Methods:

• Application of an Artificial Neural Network (ANN) training algorithm inspired by neuronal metaplasticity.
• Utilizing the AMMLP (Adaptive Metaplasticity MLP) algorithm for credit scoring.
• Testing the algorithm on the Australian and German credit data sets.

Main Results:

• The AMMLP algorithm demonstrated superior performance compared to state-of-the-art classification algorithms in credit scoring.
• The algorithm showed particular efficiency in scenarios with limited data for certain classes or when low-probability events are critical.
• Weight updating was effectively emphasized for less frequent activations, improving model sensitivity.

Conclusions:

• The proposed ANN training algorithm based on metaplasticity offers a significant advancement in credit scoring.
• AMMLP provides a more accurate and robust method for assessing credit default risk.
• This approach holds promise for improving financial risk management and decision-making.