Stock price prediction using deep learning
Date
2025
Authors
Mirzazadeh, Saeid
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
Financial markets are inherently complex, volatile, and influenced by a wide range of factors extending beyond historical price patterns. Traditional stock price prediction models, relying solely on technical or statistical methods, often fail to account for external drivers such as macroeconomic conditions and investor sentiment. This thesis addresses these limitations by proposing a hybrid deep learning framework—CNN-LSTM-ASTL—designed to enhance stock price forecasting through the integration of structured financial data, macroeconomic indicators, and unstructured sentiment data. The model leverages Convolutional Neural Networks (CNN) for spatial feature extraction, Long Short-Term Memory (LSTM) networks for capturing temporal dependencies, and an Adaptive Spatiotemporal Learning (ASTL) mechanism to dynamically adjust to changing market conditions. A comprehensive ETL pipeline was developed to automate multi-source data collection, preprocessing, and feature engineering. The system was deployed using cloud-based infrastructure to enable scalable, real-time predictions. Empirical evaluation focused on Tesla Inc. (TSLA) demonstrated that the proposed framework outperformed traditional models such as ARIMA, Random Forest, and LSTM-only architectures across key performance metrics, achieving an R2 score of 0.912 and a Directional Accuracy of 76.5This research contributes to the advancement of AI-driven financial forecasting by demonstrating the value of combining deep learning with alternative data sources in a scalable, adaptable framework. The findings highlight both the potential and limitations of such systems, emphasizing their role as decision-support tools within modern financial analytics.
Description
Keywords
electrical engineering, CNN-LSTM-ASTL