Neural networks (NNs) have successfully been applied to classification problems including credit scoring. The tuning of the structural parameters of the NNs has a direct impact on their accuracy. In this paper, a hybrid approach based on the genetic algorithm (GA) is proposed to adjust the structural parameters of a classifier NN to achieve high accuracy. Two well-known credit scoring datasets—Australian and German datasets—are used to test the proposed approach. The results indicate that the proposed hybrid approach is able to successfully tune the structural parameters, while the accuracy of classification is enhanced and its complexity dramatically diminished in comparison with other existing approaches. The performance of the proposed algorithm has been investigated through statistical analysis The best-known solutions achieved by the proposed approach have an accuracy equal to 97.78% and 87.1% for Australian and German datasets, respectively. The results indicate 2.68% and 0.1% improvement in comparison with the best results reported in the literature, respectively. This improvement is important for real cases in which millions of loans are allocated using credit scoring approaches.

中文翻译：

---

使用遗传算法调整神经网络的结构参数：信用评分应用

神经网络 (NN) 已成功应用于包括信用评分在内的分类问题。神经网络结构参数的调整对其准确性有直接影响。在本文中，提出了一种基于遗传算法（GA）的混合方法来调整分类器神经网络的结构参数以实现高精度。两个著名的信用评分数据集——澳大利亚和德国数据集——用于测试所提出的方法。结果表明，与其他现有方法相比，所提出的混合方法能够成功地调整结构参数，同时提高了分类的准确性，并且其复杂性显着降低。已通过统计分析研究了所提出算法的性能。所提出的方法获得的最著名的解决方案对澳大利亚和德国数据集的准确率分别为 97.78% 和 87.1%。结果表明，与文献报道的最佳结果相比，分别提高了 2.68% 和 0.1%。这种改进对于使用信用评分方法分配数百万笔贷款的实际情况很重要。