Paper Key : IRJ************383
Author: Seera Naveen Saikumar
Date Published: 15 Nov 2024
Abstract
Forecasting the stock price has been one of the grand challenges in the nancial sector. With the discovery of machine learning models, especially deep learning architectures, research and innovation in nancial forecasting have been nothing less than revolutionary. Over recent years, one observes that transformer architectures originally conceived for natural language processing tasks has emerged as a promising approach to time series prediction, including stock price forecasting. This paper covers the use of transformer. The model for the stock price forecast, with the self-attention mechanism, is applied because it can implement complex dependences and relations in the domain of nancial time series. Indeed, it turns out that transformers model long-range dependencies much more eciently than traditional recurrent neural networks (RNNs) or long short-term memory (LSTM) networks tends to have a problem with vanishing gradients .The method proposed is optimization of the accuracy of predic