Using Class Decomposition for Building GA with Fuzzy Rule-Based Classifiers - Developing Concepts in Applied Intelligence - page 67

Ishibuchi et al. [3] examined the performance of a fuzzy genetic-based machine
learning method for pattern classification problems with continuous attributes.
However, when GA is applied to larger-scale real-world classification problems,
it still suffers from some drawbacks such as the ineciency in searching large
solution spaces, the diculty in handling the internal interference of training
data, and the possibility of getting trapped in local optima. A natural approach
to overcome these drawbacks is to decompose the original task into several sub-
tasks based on certain techniques. The purpose of decomposition methodology
is to break down a complex problem into several manageable sub-problems while
providing attention to each subproblem independently.
According to Michie [10], finding a good decomposition is a major tactic,
both for ensuring the transparent solutions and for avoiding the combinato-
rial explosion. It is generally believed that problem decomposition can bene-
fit from conceptual simplification of the problem, making the problem more
feasible by reducing its dimensionality, achieving clearer results (more under-
standable), reducing run time by solving smaller problems and using parallel
or distributed computation, and allowing different solution techniques for indi-
vidual sub-problems. Various task decomposition methods have been proposed.
These methods can be roughly classified into the following categories: functional
modularity, domain modularity, class decomposition, and state decomposition,
according to different partition strategies [11,13].
However, most of them are used in artifcial neural networks (ANN), very few
find their applications in GA, especially GA-based classification [14]. As GAs
have been widely used as basic soft computing techniques, the exploration of
class decomposition with GAs becomes more important. Decomposition methods
have been used in various fields, such as classification, data mining, clustering,
etc. [15] presented a feature decomposition approach for improving supervised
learning tasks. The original set of features is decomposed into several subsets.
A classification model is built for each subset, and then all generated models
are combined. A greedy procedure is developed to decompose the input features
set into subsets and to build a classification model for each subset separately.
Weile and Michielssen [16] explored the application of domain decomposition
GAs to the design of frequency selective surfaces. Masulli and Valentini [17] pre-
sented a new machine learning model for classification problems. It decomposes
multi-class classification problems into sets of two-class sub-problems which are
assigned to non-linear dichotomizers. Chan and Zhsu [18] proposed a similar
approach using GA and used Fisher's linear discriminant (FLD) Algorithm to
re-assemble the rules. Apte et al. [19] presented a new measure to determine the
degree of dissimilarity between two given problems and suggested a way to search
for a strategic splitting of the feature space that identifies different characteris-
tics. Watson and Pollack [20] used techniques from multi-objective optimization
to devise an automatic problem decomposition algorithm that solves test prob-
lems effectively.
In this work we propose a new solution based on the class decomposition
approach, in which a classification problem is fully partitioned into N class