2024/10/08 by Ahang golabi, golabi, Ahang, Salehi, Mojtaba +2
Decision Sciences · Computer Science · #Stock Market Forecasting Methods #Currency Recognition and Detection #Advanced Technologies in Various Fields
paper · doi:10.71644/admt.2025.1186141
This study introduces a new method for identifying trading signals in the financial market using machine learning techniques. It employs graphical and correlational research methods, utilizing support learning techniques in MATLAB to test hypotheses. The study proposes an automated trading system combining deep learning and reinforcement learning to determine trade signals and position sizes. The framework combines an LSTM network with Q-learning, an out-of-policy reinforcement learning algorithm. Q-learning aims to maximize overall reward by learning from actions that deviate from the current policy. This study introduces a new framework that utilizes the collective intelligence of multiple expert traders to learn across different time frames. It shows that using Fundamental and technical indicators independently or in combination to train LSTMs for predicting currency movements in Forex significantly improves prediction accuracy. The study introduces a third class to represent small changes in currency pair prices between two consecutive days, improving prediction accuracy. It also describes a new method for determining the most appropriate threshold value for defining the unchanged class. Additionally, the study trains LSTMs to predict values k days into the future, and explores the impact of different training iterations on accuracy values.