Comparative Analysis of Different Machine Learning Techniques in Forecasting Stock Price

The anticipation of stock values has always been appealing and challenging for shareholders due to its inherent complexity, nonlinear nature, and dynamic changing patterns. This study focuses on predicting the future values of stocks by different models and comparing their performances. Specifically, it examines three companies: International Business Machines, General Electric, and ExxonMobil. Historical data spanning about 53 years was collected for these companies. The research generates value predictions for these groups based on different time steps: 7, 14, and 30 days in length. To make these predictions, various machine learning algorithms were employed, including artificial neural networks (ANN), autoregressive integrated moving average (ARIMA), convolutional neural network (CNN), Transformer, gated recurrent units (GRUs), and long short-term memory (LSTM). For each prediction model, the stock's five common values of different time intervals were selected as inputs. The study concludes that Long Short-term Memory outperformed other algorithms in forecasting the stock prices among these three companies, not only on time steps but also at different time intervals, exhibiting higher accuracy in capturing long-term dependencies and recurrent patterns. Moreover, under the condition of huge data sizes, there were notable improvements in the Transformer model to achieve the best results.