Evolving fuzzy ensemble : online learning fuzzy neural network with dynamically changing data

• Main Author: Soo, Wei Quan
• Other Authors: Quek Hiok Chai
• 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/10356/62790
• Institution: Nanyang Technological University

Fuzzy Neural technique is often used to model dynamic data stream in the financial market and to examine hypothetical cases. In the event of catastrophic change, old rules become irrelevant which will affect the accuracy and the performance of the network. In addition, having unnecessary rules in the rule base will increase the computational time. The aim of this project is to implement the concept of the Bienenstock- Cooper Munro (BCM) theory on evolving fuzzy ensemble (eFE), which is error-driven progressively adapted computational structure that is defined by a set of highly transparent Mamdani fuzzy rules, and to experiment and benchmark the performance and accuracy of the network with the other several networks. Through implementation of BCM theory, the problem can resolve by incorporating synaptic potentiation (association or Hebbian) and depression (dissociation or Anti-Hebbian) which is useful for time-variant data computations. These strengths are controlled in an associative and disassociative manner that will aid in developing selectivity and reducing the loss of information. The proposed method is compared against with those of the existing neuro-fuzzy architectures in both time-invariant and time variant data.