چكيده به لاتين
The prediction of stock market has become a prominent area of attention, driven by the significant wealth that investment returns in the stock market can generate. Accurate stock market prediction is a complex task in the financial domain. Decades ago, analysts used traditional techniques to predict prices based on historical data. However, due to specific characteristics of financial markets, such as volatility, anomalies, and oscillatory trends, this approach is less effective. Time series models such as ARMA, ARIMA, GARCH, and GARCH generalized have become popular among analysts and researchers. Nevertheless, due to limitations in handling the chaotic and highly volatile nature of the stock market system, researchers have turned to the use of conventional time series models, such as fuzzy models. Additionally, many intelligent techniques, such as genetic algorithms, neural networks, and backpropagation algorithms, have proven valuable for researchers in predicting financial markets. In this thesis, after reviewing previous works on stock price prediction and providing a comprehensive overview of machine learning concepts, a framework for predicting stock prices on the Tehran Stock Exchange is presented. Subsequently, the prediction of selected stocks, identified through the stock selection process, is performed using a deep learning algorithm (LSTM) and three machine learning algorithms (SVR, Random Forest, and XGBoost). The results show the superiority of XGBoost and Random Forest methods in predicting stock prices, with average accuracies of 97.80% and 97.79%, respectively, compared to other algorithms.
