Related Experiment Videos
A fuzzy min-max neural network classifier with compensatory neuron architecture.
Abhijeet V Nandedkar1, Prabir K Biswas
1Department of Electronics and Electrical Communication Engineering, Indian Institute of Technology, Kharagpur 721302, India. avnandedkar@yahoo.com
IEEE Transactions on Neural Networks
|February 7, 2007
Summary
This study introduces a fuzzy min-max neural network classifier with compensatory neurons (FMCN) for improved pattern classification. FMCN efficiently handles hyperbox overlaps, reducing errors and learning data online in a single pass.
Area of Science:
- Artificial Intelligence
- Machine Learning
- Neural Networks
Background:
- Fuzzy min-max neural networks (FMNN) use hyperbox fuzzy sets for classification.
- Existing methods like Simpson's contraction process for hyperbox overlap have limitations.
Purpose of the Study:
- To propose a novel fuzzy min-max neural network classifier with compensatory neurons (FMCN).
- To address hyperbox overlap and containment issues more efficiently than existing methods.
- To improve online learning capabilities and reduce classification errors.
Main Methods:
- Introduction of compensatory neurons (CNs) inspired by human reflex systems.
- FMCN activates CNs for test samples in overlapped regions between classes.
- Elimination of the erroneous contraction process used in prior methods.
Main Results:
- FMCN demonstrates efficient handling of hyperbox overlap and containment.
- The classifier learns data online in a single pass with reduced errors.
- Performance is robust and less dependent on the initialization of the expansion coefficient.
Conclusions:
- FMCN offers an effective alternative to traditional FMNN and GFMN classifiers.
- The compensatory neuron architecture enhances classification accuracy and efficiency.
- FMCN provides a more reliable and adaptable approach to supervised classification.
