基于模糊核超球的快速分类算法在股票预测中的应用

摘要

SVDD classification algorithm based on SVM has defects, such as high calculation complexity property and low accuracy. According to nonlinear and high-noise characteristics of stock data, inspired from the idea of traditional SVDD classification algorithm, the proposed algorithm (FCABFKH) adopts mergence method to find hypersphere sets and maximum membership degree law to construct classifier. By this means, the algorithm can rule out off-group points and hypersphere sets overlap problem. Furthermore, it can avoid complex quadratic programming. Consequently, FCABFKH provides faster rate and higer accuracy. Using the data of listed companies of China A stocks market, experiments are done to test the validity of the method mentioned above. The result indicates that portfolio's return rate using classification method of FCABFKH is higher than the market benchmark.%基于 SVM(支持向量机)的 SVDD(支持向量数据描述)分类算法存在计算复杂、分类准确率较低的缺陷，针对股票数据非线性、高噪声的特点，在传统的 SVDD 分类算法基础上，模糊核超球快速分类算法(FCABFKH)通过合并法寻找超球集，并依据最大隶属度原则构建分类器，排除了离群点和超球集的重叠问题，同时避免了复杂的二次规划，具有分类速度快，分类结果准确率高的特点。采用中国沪市上市公司数据验证该方法的有效性，实验结果表明，运用FCABFKH算法得到的组合回报率超过了市场基准。