A study of the transaction volume prediction problem based on recurrent neural networks

Abstract

With the rapid development of artificial intelligence technology, intelligent fintech scenarios based on big data are receiving more and more attention, and through the analysis of massive financial class data, accurate decision support can be provided for its various scenarios. By predicting the transaction volume of a financial product of a bank, abnormal transaction flow and gradual change trend can be found 1 day in advance to provide decision support for business department program development, and provide decision support for system expansion and contraction, thus reducing system online pressure or releasing unnecessary system resources. Linear algorithms such as AR model, MA model, ARMA model, etc. have poor prediction results for transaction volumes during holidays in the non-stationary dataset handled in this study due to strong assumptions on historical data. In this paper, we design and implement an LSTM-based trading volume prediction model LSTM-WP (LSTM-WebPredict) using deep learning algorithm, which can improve the accuracy of prediction of holiday trading volume by about 8% based on the linear algorithm by discovering and learning the features of historical data, and the learning ability of the model will gradually increase with the increasing of training data; Not only that, the research of this algorithm also provides corresponding technical accumulation for other business scenarios of time series problems, such as trend prediction and capacity assessment.