Convolutional Neural Networks Based Detection Of Outliers And Turning Points In Stock Trading

There are many frequent fluctuations and all kinds of abnormal points in the trading of stocks which makes investors confused when choosing stock trading.Moreover,In the process of stock trading,everyone want to obtain the reversal of the stock points which including the valley price and the peak price.so they can get more profits.Thus,it is important to identify the abnormal fluctuations and the reversal of the stock points.In recent years,the convolutional neural network made some achievements in terms of image mining.So we want to use CNN to identify the abnormal fluctuations and the reversal of the stock turning points.In order to capture the style of the stock turning points,we should get all the stocks data in history,and according to the K chart,label all the stock turning points by manually.We use sliding window to split the stock sequence into subsequence.Then,input the subsequence directly to the CNN to get the characteristics by automatically.The Convolution neural network consists of a number of convolution,pooling layer,and combining layer which is decided by the number of time series attributes.The output of the CNN is the characters of the subsequences.Then,input the characters into a MLP to get the Classify labels.To train the network we will use the classical algorithm forward propagation algorithm and back propagation algorithm.And last,use part of the data to measure the algorithm.To identify the abnormal fluctuations points,we use the same CNN to get the characters of stocks.And then input the characters into the Wave Cluster.according to the definition of the stock abnormal points,and then use the Markov to get the stock abnormal points as the time changes.Last measure the way to identify the stock abnormal points by the announcement from Stock Exchange.We get all the Shanghai and Shenzhen A stocks historical data from August 20(2015)to April 20(2016),and the Stock abnormal transaction announcement,at the same time,we label all the turning point by manually.Then,using those data to train the model to get the turning point and abnormal points.The results of the experiments on the stock turning points and abnormal points detection show that the convolution neural network,time series analysis method,and the classification of the traditional clustering methods of combining the model can effectively detect the inversion point and abnormal point stock information.