Extraction of rules from artificial neural networks for nonlinear regression

Area of Science:

• Artificial Intelligence
• Machine Learning
• Data Science

Background:

• Neural networks (NNs) are powerful function approximators, capable of solving complex problems without prior data distribution knowledge.
• Extracting interpretable knowledge from NNs is crucial for understanding their decision-making processes, particularly in regression.
• Existing methods for rule extraction primarily focus on classification, with limited options for regression tasks.

Purpose of the Study:

• To develop and present a novel approach for extracting symbolic rules from trained neural networks specifically for regression problems.
• To enable the explanation of how neural networks solve regression tasks by providing interpretable rule sets.

Main Methods:

• The proposed method extracts rules from trained neural networks (NNs) for regression.
• Each extracted rule defines a subregion of the input space.
• Within each subregion, the network's output is approximated by a linear function of the input attributes.

Main Results:

• The approach was evaluated on 32 benchmark datasets.
• Experimental results demonstrate the effectiveness of the method in generating accurate regression rules.
• The extracted rules provide insights into the network's behavior for regression tasks.

Conclusions:

• The presented approach successfully extracts accurate and interpretable rules from trained neural networks for regression.
• This method addresses a gap in the literature concerning rule extraction for neural network-based regression.
• The findings highlight the potential for enhancing the explainability of neural networks in regression applications.