Utilizing Deep Learning to Predict the Tehran Stock Exchange Stock Market

Abstract

For investors, the nature of stock market movement has always been confusing due to a
variety of relevant elements. With machine learning and deep learning algorithms, this study
intends to drastically minimise the risk of trend prediction. Four stock market groupings,
including diversified financials, petroleum, non-metallic minerals, and basic metals, are
selected for experimental evaluations on the Tehran stock exchange. This study compares
nine machine learning models (Decision Tree, Random Forest, Adaptive Boosting
(Adaboost), eXtreme Gradient Boosting (XGBoost), Support Vector Classifier (SVC), Nave
Bayes, K-Nearest Neighbors (KNN), Logistic Regression, and Artificial Neural Network
(ANN)) and two potent deep learning techniques (Recurrent Neural Network (RNN) and
Long short-term memory) (LSTM). 10 technical indicators derived from ten years of
historical data are our input values, which are intended to be utilised in two ways. First,
computing the indicators based on the stock trading values as continuous data, then
transforming the indicators to binary data prior to use. On the basis of the input ways, each
prediction model is assessed using three metrics. The evaluation findings reveal that for
continuous data, RNN and LSTM significantly outperform other prediction models. In
addition, the evaluation of binary data demonstrates that these deep learning approaches
perform the best; however, the performance improvement of models in the second method
reduces the gap.