Computational intelligence-based classifiers in multi-dimensional pattern recognition.

摘要

Automatic pattern recognition is becoming overwhelmingly important in a variety of areas such as medical diagnosis, engineering and economics. This multitude of data inspires development of new and powerful computational tools.; Computational intelligence (Cl) combines in synergy several advanced information processing technologies: neural networks, fuzzy sets theory and evolutionary computation. The objective of this study is to investigate three methods in which CI is involved with pattern recognition: (1) A feature extraction method using a genetic algorithm (GA). The GA serves as an optimization tool for constructing a piecewise representation of highly dimensional data. (2) A fuzzy adaptive logic network (FALN) used as a pattern classifier. The FALN realizes a topology of perceptrons combined with a structure of fuzzy neurons, which has the advantages of both processing capabilities and transparency. (3) Lastly, FALNs are incorporated in an ensemble (using bagging and boosting methods) for achieving improved accuracy.