Classification via classifier ensembles: principles, theory and applications

Pattern recognition techniques are an important component of intelligent
systems and are used for decision making, object and pattern
classification, and data preprocessing.
In particular, the classification problem requires the partition
of input data into a number of semantic classes or categories.
Depending on the application, data can be images or signal waveforms or
any type of measurements that need to be classified.
When the classification task is actually difficult two approaches are
commonly used: to design a single sophisticated classifier or to design
a classifier ensemble.
In this talk, an overview of the classifier ensemble approach will be
presented. First, the genesis of multiple classifier systems and the
main background concepts will be described. Second, the main methods and
algorithms for designing multiple classifiers systems will be presented.
In particular, combining multiple classifiers can be divided in two
categories: classifiers fusion and classifiers selection. Both
approaches will be illustrated with some examples like voting methods,
Bayesian methods, weighted methods, linear combiners, adaptive and
trained methods.
Some theoretical and empirical results comparing the classifiers
ensemble and single classifiers methods will be finally shown.