Financial prediction using type-2 fuzzy inference systems

• Main Author: Ong, Randall Qin-Yi
• Other Authors: Suresh Sundaram
• Format: Final Year Project
• Language: English
• Published: 2015
• Subjects: DRNTU::Engineering::Computer science and engineering::Computing methodologies::Artificial intelligence
• Online Access: http://hdl.handle.net/10356/62838
• Institution: Nanyang Technological University

Financial markets have long been seen as a place where fortunes are made and lost. Should an individual thread carefully with intelligent decisions, he or she could stand to potentially profit off the markets with minimal risk. With advancements in machine learning and data processing, the idea of achieving a single intelligent system that can ply the markets successfully draws nearer with each day. This project attempted to design and built a system to trade the currency markets based on an Interval Type-2 Neural Fuzzy Inference System. Neural networks are traditionally known to be black boxes where the reason behind certain learning behaviours cannot be specifically explained. Type-2 fuzzy set theory aims to solve this by combining the learning capability of a neural network with the categorization of fuzzy set theory. The learning in this project is selective and based on the learning parameters for error and uniqueness to delete, learn or reserve a data sample. In the 6 layer neural network, financial indicators were used together with the Open-High-Low-Close of correlated currency pairs as data inputs. The performance of learning was highly satisfactory although the overall performance for actual prediction was not great. Potential solutions to improve the prediction performance of the system would be to include multiple different and independent prediction systems and a way to accurately include textual data in the form of economic reports and new releases. Some recommendations for the project include a natural language processing module, a live data feed with financial news providers like Bloomberg, and a larger data set to cover even more ground with attempting to find hidden correlations.