Generic self-evolving TSK fuzzy neural network with rough set (GSETSK+RS)
Fuzzy neural networks are the hybrid of artificial neural networks and the fuzzy systems. The combination unites the strengths and eliminates the weaknesses of the individual system. There are two types of fuzzy neural networks: the Mamdani model and the Takagi-Sugeno-Kang (TSK) model. For complex p...
Saved in:
Main Author: | |
---|---|
Other Authors: | |
Format: | Final Year Project |
Language: | English |
Published: |
Nanyang Technological University
2021
|
Subjects: | |
Online Access: | https://hdl.handle.net/10356/149161 |
Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
Institution: | Nanyang Technological University |
Language: | English |
Summary: | Fuzzy neural networks are the hybrid of artificial neural networks and the fuzzy systems. The combination unites the strengths and eliminates the weaknesses of the individual system. There are two types of fuzzy neural networks: the Mamdani model and the Takagi-Sugeno-Kang (TSK) model. For complex problems that require high precision, the TSK model is preferred over the Mamdani model for its higher modelling accuracy.
Generic self-evolving TSK fuzzy neural network (GSETSK) was proposed to address the existing problems of TSK networks. Equipped with Multidimensional-Scaling Growing Clustering (MSGC) and ‘gradual’-forgetting-based rule pruning, GSETSK can learn in an incremental manner, unlearn outdated data and maintain a compact and interpretable rule base. However, it is unable to maintain an interpretable rule base for large dimensional problems where more terms and rules are required to represent the additional features. This issue can address with feature reduction.
This paper proposes GSETSK with rough set (GSETSK+RS) where rough set-based attribute reduction can be performed to reduce the number of attributes and consequently remove the duplicated rules. Rough set theory which is typically applied to the Mamdani model is adapted for the TSK network.
The proposed GSETSK+RS is experimented on real-world problems including cancer detection and stock price forecasting. The performance is benchmarked against GSETSK and other fuzzy neural networks. GSETSK+RS is also applied in a trading system to determine its applicability for trading. The overall results are promising. |
---|