Stock Price Prediction Using Optimized Generative Adversarial Network (Gan) with Attention Bidirectional Gated Recurrent Unit and Convolution Neural Networks (Attbigru-Cnn)

Abstract

Forecasting stock price variations is one of the important problems for economists. A robust and precise forecast substantially mitigates investment risks for stakeholders. Currently, the fastest growth in Artificial Intelligence (AI) and Deep Learning (DL) models shed light on the intricacies of Stock price prediction (SPP). While DL models has demonstrated remarkable efficacy in long term SPP, it does handle the challenges such as intricate model structures, large training processes, substantial computational demands, and high cost of utilization. This study proposes a SPP model using optimized Generative Adversarial Network (GAN). It consists of Attention Bidirectional Gated Recurrent Unit (AttBiGRU) model as a discriminator that separates the generated stock price from the true stock price, and as a generator that uses past stock price inputs to produce future stock prices. Moreover, the hyperparameters of GAN are optimized by applying the Walrus Optimization Algorithm (WaOA). The proposed GAN:AttBIGRU-CNN model has been applied over 4 bench mark datasets. Experimental results state that GAN:AttBIGRU-CNN model attains minimized statistical error measures while predicting the stock prices, when compared to the existing models.