Predicting the stock market's closing price accurately and reliably is complicated. By learning from past market behavior, deep learning algorithms can potentially forecast future price movements with some accuracy. Large volumes of historical stock data can be analyzed by deep learning (DL) to find trends and correlations that support predictive modeling. Deep learning algorithms may be able to predict future price changes with a certain level of accuracy by learning from historical market activity. Autoencoders (AE) retrieve features that recognize important data patterns. The study offers a Gated Recurrent Unit (GRU) model with intricacies in Stock market prediction, such as market volatility and complicated patterns for the model’s performance. A single dataset's impressive performance does not establish the model's durability. The proposed model is split into 80% training data,10% testing, and 10% validation data The study justifies the model's stability and extensibility by doing a thorough comparison analysis on 30 datasets of companies' stocks, accomplishing an average Mean Squared Error (MSE) of 0.001, Mean Absolute Error (MAE) of 0.021, and Max error of 0. 213. In addition to our novel strategy, we retrained previously employed deep learning models using the same dataset for validation to evaluate our suggested model's contribution. Furthermore, we contribute this offers vital information for improving future Stock market prediction models by revealing the distinct contributions of each component to determine the stock closing price.