An improved technique for stock price prediction on real-time exploiting stream processing and deep learning

• Published in: Multimedia tools and applications 2023-12, Vol.83 (19), p.57269-57289
• Main Authors: Bandhu, Kailash Chandra, Litoriya, Ratnesh, Jain, Anshita, Shukla, Anand Vardhan, Vaidya, Swati
• Format: Article
• Language: English
• Subjects: Accuracy; Cleaning; Computer Communication Networks; Computer Science; Data collection; Data Structures and Information Theory; Deep learning; Forecasting; Industrial applications; Multimedia Information Systems; Root-mean-square errors; Special Purpose and Application-Based Systems; Stock prices
• ISSN: 1573-7721; 1380-7501; 1573-7721
• DOI: 10.1007/s11042-023-17130-x

The proposed model is a Deep Learning (DL) based method employing Long Short-Term Memory (LSTM) networks for forecasting stocks. The aim of this approach is forecasting stock prices of Apple Inc. using statistics on previous stock prices obtained from Tiingo. The proposed model consists of several stages of processing and modelling, including data cleaning, feature selection, feature scaling, model building, model evaluation, model improvement, and prediction. Cleaning, organising, and transforming raw data into a format appropriate for analysis are all parts of data pre-processing. Feature engineering involves the data extraction and selection of relevant features for accuracy improvement of the model. The scaling of features involves normalising the data to prevent bias in the model. The LSTM models are built and evaluated using multiple metrics such as Mean Squared Error (MAE) and Root Mean Squared Error (RMSE). The model is iteratively improved using a combination of hyperparameter tuning and feature engineering. Finally, the model is then used to forecast stock prices for the following 30 days, and the accuracy of the forecasts is determined. The proposed methodology is designed to outperform traditional LSTM models for predicting the future price of stock by incorporating novel techniques, for feature engineering and model refinement. The suggested design is a comprehensive approach for forecasting future stock prices using DL based techniques. The model is designed to be flexible and adaptable, allowing for customization for different datasets and prediction horizons. It represents a significant improvement over existing LSTM models for stock price prediction to be valuable in a variety of financial industry applications. This paper collects data from Tiingo API and uses stacked LSTM to train the model. The experimental results give only 0.0813 RMSE, which proves that the model is more accurate and precise.