Research on the Stock prediction based on LSTM Algorithm

Abstract

Predicting the value of a stock in the future is a widely recognized challenge. However, the Long Short-Term Memory (LSTM) network provides a distinctive solution as a specialized type of recurrent neural network proficient at tackling long-term dependency problems. Each time a new input is fed, the LSTM will first decide which previous memories to forget based on the new input and the output of the previous moment. This paper analyzes the LSTM algorithm for time series prediction in volume of stock. The dataset is initially split in a 7:3 ratio, with 70% allocated for the training set and 30% for the test set. This paper conducts data processing first, and then plots time series graphs to visualize the characteristics of volume of stock. Additionally, this paper conducts smoothness test together with white noise test, and makes time series predictions using LSTM. Values obtained from the experiments show that the training MSE of the LSTM algorithm is 28333951.7500 and the RMSE is 4722325.29. The final prediction comparison graph proves the potential of the LSTM algorithm to be used in market prediction.