基于最小方差的股市拐点预测方法

发布时间：2018-02-28 19:04

本文关键词： 股市拐点预测不平衡分类最小方差法 SVM 惩罚因子　出处：《计算机应用研究》2017年11期 　论文类型：期刊论文

【摘要】：预测股市拐点对指导股市投资具有重要意义,而股市拐点的预测是不平衡分类问题。针对SVM在解决此类不平衡问题时存在的偏斜性问题,提出了一种选取惩罚因子的方法。该方法以训练集交叉验证后所有类别样本的查全率与查准率的方差的乘积作为判断标准,将最小方差乘积对应的惩罚因子作为各类的最优惩罚因子,并将此Biased-SVM模型应用于股市拐点预测。在实验中,选取了常用股票技术指标作为输入向量,与其他几种解决不平衡问题的方法进行了比较。实验结果表明,最小方差法在保证非拐点的多数类样本识别精度的同时提高了两类拐点样本的识别精度,为投资决策提供了佐证与帮助。
[Abstract]:Predicting the inflection point of the stock market is of great significance to guide the investment of the stock market, but the prediction of the inflection point of the stock market is an unbalanced classification problem. In this paper, a method of selecting penalty factors is proposed, which is based on the product of the recall rate and the variance of the recall rate of all kinds of samples after cross-validation of the training set. The penalty factor corresponding to the product of minimum variance is used as the optimal penalty factor, and the Biased-SVM model is applied to predict the inflection point of the stock market. In the experiment, the commonly used technical indexes of stock are selected as input vectors. The experimental results show that the minimum variance method not only ensures the recognition accuracy of most classes of non-inflection points, but also improves the recognition accuracy of two kinds of inflection point samples. It provides evidence and help for investment decision.
【作者单位】：河北工业大学计算机科学与软件学院;河北工业大学河北省大数据计算重点实验室;河北工业大学经济管理学院;
【基金】：天津市应用基础与前沿技术研究计划重点项目(14JCZDJC31600) 河北省自然科学基金资助项目(F2016202144) 河北省高等学校科学技术研究重点资助项目(ZD2014030) 河北省科技计划资助项目(13456243)
【分类号】：F832.51;TP18

【相似文献】