An approach to development of an ensemble classification system

Generally, information system handles huge volume of dataset. Classifiers provide poor performance when such dataset are feed into it for categorization due to their high dimension. The most important attributes are extracted from the dataset prior to classification for efficient classifier design. It is also true that there may be many classifiers of a particular system, some provide better accuracy than others but selection of single classifier with all its optimized parameters is always not a good choice for various real world applications. The paper proposes a novel method for construction of an ensemble classifier by combining multiple classifiers obtained using Rough Set Theory and Genetic algorithm. The method selects the classifiers for integration based on accuracy and develops an efficient and effective ensemble classification system. In the first phase of the work, rule based classifiers termed as base classifiers are constructed from the reduced information sub systems obtained using Rough Set Theory, where a set of high quality rules are generated for each sub system. In the second phase, base classifiers are combined and an optimal ensemble classification system is developed using Genetic Algorithm. Here, ensemble classifier takes an important role to discover the class labels of the test objects with higher accuracy. The proposed algorithm has been run on standard benchmark dataset collected from the UCI repository.