Stock price prediction and comparative analysis using RNN, LSTM and ARIMA

With the rapid development and complexity of global financial market, stock market prediction has become a core issue in financial research, which plays a vital role in guiding investors’ decision-making and maintaining market stability. Traditional forecasting methods are limited by their ability to handle complex market dynamics, while the rise of data science and artificial intelligence technologies provides new perspectives and tools for stock price forecasting. Considering the nonlinear nature of stock data and the long-term dependence of time series, this study aims to explore and evaluate the validity and accuracy of three different models in stock price prediction: recurrent neural network (RNN), long short-term memory network (LSTM) and autoregressive integrated moving average model (ARIMA). We first explains the theoretical basis of RNN, LSTM and ARIMA models, and describes in detail their unique advantages in processing time series data. Then, the paper introduces the data sets in detail and makes correlation analysis on them, and introduces the hyperparameters and evaluation indexes of the stock price prediction system using these three models. In the experiment part, the model is trained and tested with real stock market data to quantify the prediction ability of the model, and the prediction performance of the three models is compared according to the experimental results. The experimental results show that the RNN model performed best on the training set, but fell short in capturing short-term stock price movements, the LSTM model had the largest prediction error on the training set, and the ARIMA model struck a balance between the two, with an error index in between.