Proposal and Implementation of Machine Learning Models for Stock Markets using Web Data

• Eduardo Jabbur Machado CEFET-MG
• Adriano César Machado Pereira UFMG

Resumo

The investment market has been growing every day, performing an important role in the lives of individuals and corporations. Therefore, there is a need to better understand the situations that occur in the capital market, by means of strategies and indicators that can assist in pattern recognition, analisys and investiment decisions. This work performs a study of characterization and analysis of a historical time series data of 10 asset codes (i.e., BBAS3, USIM5, PETR4, JBSS3, KROT3, LAME4, MRVE4, NATU3, RADL3 e TIMP3) of the Bovespa index and sentiment analysis of polarity news and Twitter data with the proposal of evaluating a prediction model. It proposes the combination of deep learning and machine learning computational intelligence models for prediction, allowing the execution and cancellation of buy and sell orders. Finally, it evaluates the behavior of each proposed trading strategy by Accuracy, Percentage of Financial Return and other indicators to provide a better understanding of financial market behavior.

Palavras-chave: Web 2.0, Data Characterization, Stock Markets, Trading Strategies, Financial Indicators, Machine Learning, Deep Learning