Financial Prices Prediction of Stock Market using Supervised Machine Learning Models

Abstract

The process of predicting stock market movements may initially appear to be non-statistical due to the multitude of factors involved. However, machine learning techniques can be utilized to establish connections between past and present data, enabling the training of machines to make accurate assumptions based on the information. By effectively linking historical data to current data using machine learning, it becomes possible to make precise predictions regarding stock performance. These predictions can lead to substantial profits for individuals and their brokers. Traditionally, stock market predictions have exhibited chaotic patterns rather than randomness, which is why a thorough analysis of a market's historical data allows for predictions to be made. Machine learning offers an efficient means of modeling such processes. By closely aligning market predictions with actual values, the analysis's accuracy can be raised greatly. The field of stock prediction has seen a growing interest in machine learning among researchers due to its effectiveness and precision. Regression-based models are commonly employed when the objective is to forecast continuous values based on independent variables. To predict stock market closing prices for the upcoming ten to fifteen days, we used SVR, RF, KNN, LSTM, GRU, and LSTM with GRU in our study. In regression modeling, the R-squared (R2) value represents the percentage of difference explained by the independent variable(s). A higher (R2) value near to 1 indicates better performance. Our experiments yielded R2 values of 0.832, 0.832, 0.574, 0.838, 0.825, and 0.815 for SVR, RF, KNN, LSTM, GRU, and LSTM with GRU, respectively. Consequently, the most effective model for correctly predicting stock market closing prices is the LSTM learning model, which had the greatest R2 value of 0.838.