A Pruning Neural Network Model in Credit Classification Analysis.

Area of Science:

• Computational Intelligence
• Machine Learning
• Financial Modeling

Background:

• Credit classification is crucial for financial decision-making.
• Artificial neural networks (ANNs) are established methods in credit scoring.
• Existing ANNs can be computationally intensive and may contain superfluous parameters.

Purpose of the Study:

• To introduce a novel pruning neural network (PNN) for credit classification.
• To enhance model efficiency and accuracy by mimicking biological neural structures.
• To enable effective hardware implementation of the proposed model.

Main Methods:

• The proposed PNN is inspired by synaptic nonlinearity in biological dendritic trees.
• The network is trained using the error back-propagation algorithm.
• Neuronal pruning is achieved by removing superfluous synapses and dendrites, forming a tidy dendritic morphology.
• Logic circuits (LCs) are utilized to simulate and implement the dendritic structures.

Main Results:

• The PNN demonstrated superior performance on the Australian and Japanese credit datasets.
• The model achieved higher accuracy compared to classical credit classification algorithms.
• PNN exhibited significant improvements in computational efficiency.

Conclusions:

• The pruning neural network (PNN) offers a more accurate and efficient approach to credit classification.
• The biologically inspired pruning mechanism effectively reduces model complexity.
• The successful simulation using logic circuits facilitates effective hardware implementation for practical applications.