Extreme learning machine based stock prediction with information theory, genetic algorithm and indicator voting mechanism

• Main Author: Gu, Yi.
• Other Authors: Ma Maode
• Format: Final Year Project
• Language: English
• Published: 2011
• Subjects: DRNTU::Engineering::Electrical and electronic engineering::Computer hardware, software and systems
• Online Access: http://hdl.handle.net/10356/45862
• Institution: Nanyang Technological University

Stock market is one of the most lucrative markets in the world. As such, it has been the center of attraction for researchers and practitioners. So far, some neural network models, such as BP and SVM, have been applied to stock prediction. However, they are either too slow or easy to converge to local optimum, which affects prediction performance. To overcome these limitations, Extreme Learning Machine is studied and applied. The fast speed and high accurate performance relative to SVM proved ELM’s effectiveness and efficiency on time series prediction. To select an optimal set of input variables for ELM, Information Theory and Genetic Algorithm are developed to select a set of optimal input features, by maximizing the relevance between input features and output targets, and minimizing the redundancy between input features themselves, the ELM performance on stock prediction is maximized. Moreover, an Indicator Voting Mechanism is proposed to make the system more robust. Thus, by integrating ELM, Information Theory, Genetic Algorithm and Indicator Voting System, the Stock Prediction System is developed. The experimental results on several stock prediction problems have shown that the system can produce effective recommendations and increase investors’ cumulative wealth by more than market average return.