سيدمهدي ناجي اصفهاني

Success in stock trading relies on making accurate and timely decisions for entering and exiting positions. However, due to the impact of various known and unknown factors on stock prices and their unpredictable behavior, making precise and correct decisions in this field is often challenging and requires extensive expertise and knowledge. Therefore, there is a strong need for an automated approach to effectively and efficiently utilize financial data for supporting investment decisions. Automated financial models, particularly deep learning models, have been the focus of researchers for several decades due to their performance. In this study, we aimed to enhance decision-making in stock trading by utilizing different technical analysis indicators calculated over various time intervals and transforming them into visual representations as input for the models. We developed three distinct two or three-dimensional models using convolutional neural networks to predict stock trends. Subsequently, we combined the results using ensemble learning algorithms to achieve improved accuracy and reduced variance in determining whether to buy, sell, or hold stocks for future periods. Based on eva‎luations conducted on our proposed model, it exhibited superior financial performance and accuracy compared to the related models mentioned in the literature. However, it is important to note that the Iranian stock market and securities market present certain limitations, including one-sided market conditions, high transaction costs, and the manual labeling of data. Consequently, the financial performance of our model was relatively comparable to a basic buy-and-hold strategy considering these market constraints.

يادگيري عميق , يادگيري جمعي , شبكه عصبي پيچشي , تحليل سهام , بورس اوراق بهادار

Deep learning , Ensemble learning , Convolutional Neural Network , Stock Exchange , Stock analysis

Mehdi Naji

Dr.Mehdi Ghazanfari