A Stock Ranking Method Based On Machine Learning

Over the years,people always want to know the rules behind the operation in the stock market financial investment and make analysis and forecasting on it.The investment experts from different countries also use the massive stock data for data mining through the use of different investment analysis methods in order to find out the potential rules of operation and the rules of stock exchanges behind the stock market and make a forecast on the change of the stock market development in the future to achieve the purpose of maximize revenue.The main content of this paper is the change of the stock data of listed companies and the stock price.According to the change of stock price during the research period and the calculation of an eigenvalue,an optimized version of k-nearest neighbor algorithm is designed.Then,an upward trend system model is established to forecast the stock price trend of the listed company,and select the suitable listed companies for their investment by their own risk type.With the large data-related technologies are more and more mature,in choosing a platform of large-scale stock data set,the main consideration in this paper is the memory consumption and data computing efficiency,which uses the HDFS distributed file system and more efficient distributed computing framework MapReduce on the large data Hadoop,making the ETL process operation of the entire data set can be efficient and convenient.Machine learning is also a key problem in this paper.After a deep study on the KNN algorithm,a recognition algorithm on large data model is proposed.In addition,this paper presents three different feature sets,minute price feature,K-line feature,and equity feature.After the experiment on a large number of real stock data,it is indicated that all kinds of feature sets are effective for the predicting the trend of stock price.The prediction results obtained by the recognition algorithm on large data on the same feature set are better than k-nearest neighbor algorithm.And on the different kinds of feature sets,the accuracy rate of the forecasting results on the equity feature set is much higher than that of the minute price feature set and the K-line feature.The research of this paper provides an effective method on selecting the appropriate trading objects in a large number of stocks.