AI-Driven Stock Price Prediction: A Case Study on Agricultural Development Bank Limited (ADBL) using LSTM, GRU and TCN with and without News-Headling Sentiment

Abstract

Stock price prediction in emerging markets such as Nepal presents significant challenges due to high market volatility, limited liquidity, and the strong influence of investor sentiment. This study presents a comparative analysis of three deep learning architectures Long Short-Term Memory (LSTM), Gated Recurrent Unit (GRU), and Temporal Convolutional Networks (TCN) for stock price prediction using historical OHLCV data from the Nepalese capital market. To assess the contribution of market sentiment, financial news headlines are incorporated and transformed into sentiment scores using the FinBERT model. Model performance is evaluated under two experimental settings, with and without sentiment features, using Mean Squared Error (MSE), Root Mean Squared Error (RMSE), Mean Absolute Error (MAE),and R-squared (R2).The empirical results demonstrate that the GRU model consistently outperforms LSTM and TCN, exhibiting superior predictive accuracy and robustness. Furthermore, sentiment integration enhances the performance of the GRU model, while its impact varies across other architectures. These findings underscore the effectiveness of sentiment-aware deep learning models for stock price forecasting and provide valuable insights for AI-driven decision-making in emerging financial markets such as Nepal.