计算机工程与应用2025，Vol.61Issue(18)：157-165,9.DOI:10.3778/j.issn.1002-8331.2406-0062
基于RL-Net改进的可解释规则学习方法
Improved Interpretable Rule Learning Method Based on RL-Net
摘要
Abstract
As the demand for interpretable,transparent,and effective models continues to grow,rule-based classification models have become a research focus due to their intuitive understanding and good performance.In view of the fact that many existing models are limited by difficulties in structural optimization or rely on specific heuristic search strategies,a network structure SRL-Net based on RL-Net is proposed.This model learns rules through neural networks,introduces attention mechanism and pruning layer into the network,and aims to improve the accuracy and simplicity of learning rules while improving model performance,reducing unnecessary rules,and using the pruning"mask"to realize the sec-ondary refinement of rules to obtain a concise list of explainable rules.SRL-Net is experimentally verified on 12 datasets.The results show that SRL-Net has good performance on datasets of different sizes.Compared with the other 8 models,SRL-Net achieves the highest accuracy in 8 datasets and the highest value in 9 datasets.The rule complexity is reduced by about 50%on average compared with RL-Net.The experiment show that SRL-Net is an effective interpretable rule learning method.关键词
可解释性/规则学习/神经网络Key words
interpretability/rule learning/neural networks分类
信息技术与安全科学引用本文复制引用
张道胜,蒙祖强..基于RL-Net改进的可解释规则学习方法[J].计算机工程与应用,2025,61(18):157-165,9.基金项目
国家自然科学基金(62266004). （62266004）
