| The stock market is receiving more and more attention in today’s economic life.Its trend is affected by various factors such as economy,politics,and culture,and its operation law is complicated.Accurately grasping the laws of the market helps traders to make profits.In contrast,investors prefer to find stock price reversal patterns,so mining and analysis of Candlestick lines reversal patterns is particularly important.This paper takes stock historical data as the research object.Firstly,for the linear segmentation function can not effectively divide the Candlestick lines sequence,a K-line region division algorithm is proposed.On the basis of the Candlestick lines region division algorithm,it is found that the Candlestick lines pattern mining algorithm is computationally complex,so it is A mining model based on fuzzy Candlestick lines reversal pattern was found.It was further found that the pattern classification algorithm in the mining model based on fuzzy Candlestick lines reversal pattern had weak generalization ability and could not mine the K-line reversal pattern effectively.Neural network classification process and experimental verification with stock historical data.The main work of the thesis is as follows:(1)For the commonly used time series segmentation algorithms,fitting and peaking are used as the standard for dividing the sequence.In the Candlestick lines sequence,the Candlestick lines sequence cannot be effectively divided according to the trend.1.On the basis of the downward trend,a support vector machine is introduced to classify and segment the four prices in the K-line sequence.At the same time,according to the characteristics of the Candlestick lines sequence transaction,a penalty factor is added to adjust the calculation accuracy of the support vector machine.The recognition accuracy of the region segmentation method is 69.86% on average,which is 31.32% higher than the time series segmentation algorithm based on important points,and 22.89% higher than the piecewise linear single point labeling method.(2)In the existing research,the prediction model mainly focuses on the prediction of the future price.Compared with the price prediction,investors are more concerned about the reverse trend in the stock market.Therefore,on the basis of K-line sequence region segmentation,a mining mechanism based on fuzzy Candlestick lines inversion mode is constructed.The fuzzy theory is applied to the traditional Candlestick lines graph theory,and the inversion mode and inversion point are defined.The Candlestick lines morphological features are blurred,and the fuzzy features are extracted to classify the patterns.The SSE A-share data set and SZSE A-share data set were used for verification,and the three-day and five-day reversal patterns were classified respectively,with an accuracy rate exceeding 75%,which proved the feasibility of the mechanism.At the same time,the overall rate of return reached 54.43%.(3)For the pattern classification algorithm based on fuzzy Candlestick lines inversion pattern mining mechanism,the traditional classification algorithm is used to identify weak generalization ability,and good classification results cannot be obtained in unknown samples.A radial basis neural network based on local generalization error is proposed.Pattern classification model(CS-RBFNN).The model is based on the local generalization error model,and combines the cost-sensitive model with the local generalization error model to make it suitable for Candlestick lines inversion pattern classification mining.In the calculation of misclassified samples,the quasi-Monte Carlo method is used to approximate the sample distribution.Then,RBFNN is trained to obtain the optimal structure.Finally,a difference-one feature selection algorithm is proposed,which can select the number of deleted features and minimize the cost loss while deleting features.It has been verified that the CS-RBFNN local generalization error value is lower than the L-GEM algorithm and the L-GEM-WA algorithm,indicating that the CS-RBFNN generalization ability is strong.At the same time,the three-day and five-day reversal models were classified using the SSE A-share training set and the SZSE A-share training set,respectively,with an accuracy rate of 89.75%and an overall return rate of 74.45%.Compared with the mining mechanism,the accuracy rate is increased by 12.17%,and the overall return rate is increased by 20.01%.The experimental results show that the pattern classification algorithm and the mining mechanism of fuzzy K-line reversal pattern can provide effective decision support for investors. |
