Previsão financeira com aprendizado profundo e inteligência artificial

Abstract

This study is based on technological advancements and explores one of the most prominent current topics: artificial intelligence. The objective is to apply machine learning models, specifically recurrent neural networks, to forecast stock prices in the Brazilian financial market using Ibovespa data. Tests and training were conducted with different neural network architectures in search of a more accurate model. Data were collected from the Yahoo Finance API and preprocessed for model training. After implementing the architectures, the predictions were compared with actual values. Several tests were carried out by varying parameters such as learning rate, number of neurons, and other hyperparameters to identify performance improvements. The analysis of the results allowed the evaluation of the impact of these modifications and the adjustment of configurations to achieve more accurate forecasts. The final results highlight performance gains through the implementation of a hybrid neural network model, which combines different architectures to enhance predictive power and generalization. This study reinforces the potential of hybrid neural networks in solving complex computational problems and contributes to innovation in the financial context and the field of Computer Science.