Prediction of Stock Price Model through the Implementation of Hybrid BiLSTM-GRU

In the realm of finance, stock price prediction has long been a source of intrigue and significant inquiry. Investors, traders, and financial institutions are interested in having the ability to predict future stock prices accurately in order to make well-informed decisions about purchasing, disposing of, or holding onto stocks. The traditional stock prediction method is a difficult assignment that is influenced by a wide range of elements, such as investor behavior, news emotions, economic indicators, corporate fundamentals, and market trends. This study suggests using hybrid Bidirectional Long Short-Term Memory Network (BiLSTM) and Gated Recurrent Unit (GRU) to create a stock price prediction model in order to address these problems. In order to guarantee the integrity and dependability of the dataset, strong approaches for data cleaning and handling missing values are used during the initial stage of data preprocessing. The next step is feature engineering, which aims to improve the model's capacity to identify complex trends and patterns present in the dynamics of the stock market by removing pertinent elements from the raw data. The train-test split is a critical stage in accurately assessing model performance. By assuring uniform scaling across characteristics, methods of normalization are used to normalize the data and enable optimal model performance. Ultimately, two categorization models BILSTM and GRU are put into practice to fully utilize deep learning's capacity to model sequential data. In order to forecast changes in stock prices, those models are trained using preprocessed and designed features. Python is being used to implement this project. The comparison value of the proposed neural network accuracy as 92% and specificity as 91% is obtained.