• Version
• Download 1
• File Size 476.57 KB
• File Count 1
• Create Date 21 July 2025
• Last Updated 21 July 2025

Stock Market Forecasting Using an Integrated Neural Network Strategy with Feature Engineering

Dr.Nagaratna P Hegde , Dr.Sireesha Vikkurty , Cheedalla Rahul ,Polamolu Manikanta Sai Saran.

Vasavi college of Engineering,Hyderabad,Telangana,India . nagaratnaph@staff.vce.ac.in,

v.sireesha@staff.vce.ac.in, rahulcheedalla73@gmail.com, Saisaranpolamolu7136@gmail.com.

Abstract—Stock index closing price prediction remains a difficult task due to the non-linear and volatile nature of financial time series data. This study suggests a Hybrid Deep Learning with Feature Engineering (HDLFE) model that combines Long Short-Term Memory (LSTM) networks with dense layers to improve the accuracy of predictions.

The architecture is designed with two stacked LSTM layers and two dense layers, with dropout layers in between to prevent overfitting— one of the major pitfalls in conventional prediction models. The HDLFE model is trained and validated on historical closing price data of three major global indices: Nifty 50 (India), S&P 500 (USA), and Nikkei 225 (Japan).

Advanced feature engineering techniques, including normalization, are used to stabilize training and improve performance. The model performs well, with an average R-squared value of 0.997052, Mean Squared Error (MSE) of 0.000160, and Mean Absolute Error (MAE) of 0.007884. These results show the superiority of the model in detecting complex market behavior and time-based trends in different financial markets. The model's consistent performance on international indices also establishes its strength and versatility. This study contributes to the practice of financial forecasting using deep learning and provides a starting point for future improvements in stock market prediction.

Keywords—Stock prediction, Deep learning, LSTM, Hybrid model,

Feature engineering