CUDA parallel computing framework for stock market prediction using K-means clustering

Abstract

With the recent advancement of machine learning algorithms over past two decades, it has gained a lot of attention of academicians and researchers community and also found prominent application in multiple domain including finance. Our proposed model uses Machine Learning algorithms along with Parallel Computing processing to provide a new trading technique for non-stationary and multidimensional financial data of Reliance Industries retrieved from Dematerialized account from up STOX Application Programming Interface (API) which extract data at regular interval of 10 minutes. This proposed model uses K-means machine learning technique to models the gathered stock data and predicts the upcoming stock values well in advance with parallel processing techniques. This paper provides the analysis of previous year’s stock market pricing data and interpret results after performing intensive training based on machine learning algorithm on Compute Unified Device Architecture (CUDA) considering the time constraint of real time trading. The performance of the system is improved drastically with the help of machine learning techniques and to accelerate the process of generating the results, a technique of parallel computing is used in this paper. The performance time is significantly reduced because of high performance speed using CUDA parallel computing technology compared to traditional methods of single Central Processing Unit (CPU). It helped in reduction of calculation time by large margin and hence to gain book profit which is the ultimate goal of trading by predicting the stock values well in advance. Investors can decide whether to keep that stock, sell it or buy some other new stocks or neutral decision on basis of 3 clusters as predicted k means algorithm. Neutral decision means if user owns stock then he should hold that stock with him and if he does not possess stock then he should not buy it. This proposed method is suitable for intraday trading as the stocks value are taken for each 10 minutes on the basis of that model can take decision what should do.