The financial industry grapples with the rising complexity and dynamic nature of stock markets, predicting trends and risk management are key components to maximize profits and reduce losses. Artificial Intelligence have been utilized in many industries successfully, numerous efforts have been made to integrate AI into algorithmic trading. Notably, Reinforcement Learning, which can adapt to dynamic decision making. This study utilizes a deep reinforcement learning model–Deep Q-Network and assess their potential for stock market movement estimation and utilizing prediction probabilities as signal strengths and risk management. 61 candlestick patterns as well as percent change of six features open, high, low, close, volume, and adjusted close are incorporated into this process. Together with the near and volume measurements, we use integrated gradients to identify the top 10 features that have the highest integrated gradient values, yielding a total of 12 features. These are fed into a GRU (Gated Recurrent Unit) model, and the output is then mixed with holding and gain percentage information from a dense layer. Following concatenation, the output is run through more dense layers. Using a softmax function to produce probabilities for the buy, sell, and hold actions. The reward is computed using the difference between the starting amount and the total value of holdings plus the remaining balance, and actions are carried out according to the highest probability. Our method, which has been validated on American and Indian stocks, shows that the DQN model produces trading signals that are both profitable and highly effective.