An Integrating Analysis of Gan Variants with Bahdanau Attention Model using Stock Price Data

Abstract

In this research, we investigate stock price prediction using agentic Artificial Intelligence (AI) and Explainable Artificial Intelligence (XAI) frameworks, along with state-of-the-art models namely Gated Recurrent Unit (GRU), Long Short-Term Memory (LSTM), Generative Adversarial Network (GAN), and the newly designed Bahdanau Generator model. We use Bombay Stock Exchange (BSE) and Tata Steel stock datasets for the study. The datasets are pre-processed using moving averages, Moving Average Convergence Divergence (MAMACD), Fourier transform, and other technical tools. The Bahdanau Generator, specifically designed for time-series forecasting, is used to improve prediction performance.An agentic AI helps to autonomously optimize model and hyperparameter for model training while the explainable Artificial intelligence enables the enhancement of interpretability and transparency for forecast by highlighting the potential decision factor behind each financial analysis forecast. With regards to the RMSE, GRU, LSTM, GAN and Bahdanau models yield to acceptable predicted results while a higher predictive accuracy is yielded by the Bahdanau Generator.