Abstract:

Due to the complexity and particularity of financial market, the performance of existing forecasting algorithms is quite different in bull and bear markets, which results in low classification accuracy and low risk resistance. A classification algorithm based on SOM-Type-2 Fuzzy Logic System（SOM-T2 FLS） is proposed. Firstly, the samples are divided into two different subsets by SOM. Then, the T2 FLS classifier is learned under each subset. During the learning of T2 FLS classifier, the length of rule base is proposed as a regular term to reduce the complexity of the model. Finally, using the historical data of Chinese stock market, it is verified that the algorithm has better classification performance and anti-risk ability than the existing algorithms.

Key words: self-organizing feature map, Type-2 Fuzzy Logic System（T2 FLS）, rule base, model complexity

摘要：

由于金融市场的复杂性和特殊性，现有预测算法的性能在牛市和熊市中呈现出较大差异，导致分类精度不高和抗风险能力不强，提出一种基于自组织特征映射-2型模糊逻辑系统（SOM-Type-2 Fuzzy Logic System，SOM-T2 FLS）的分类算法。通过SOM网络将样本集分成两个不同子集，然后在每一个子集下分别学习T2 FLS分类器。在分类器学习过程中，提出将规则库的长度作为正则项，降低模型复杂度。利用中国证券市场的历史数据，验证了该算法较现有算法具有更好的预测效果和抗风险能力。

关键词: 自组织特征映射, 2型模糊逻辑系统, 规则库, 模型复杂度