Parametric Decomposition of Sample Space for Classification
S. Lal,, Parag A. Kulkarni, Akansha Singh · Journal of Intelligent Systems · 2010
In classification task training, sample size and high feature dimensionality enhance the computational complexity of algorithms.Therefore, many algorithms tend to work with reduced feature set and hence ignore contribution of all relevant parameters.This paper presents an Incremental, Simple, Efficient and Accurate (I-SEA) algorithm to consider contribution of all available relevant parameters while keeping the computational complexity and accuracy within acceptable limits.I-SEA partitions the training sample space during the preprocessing stage as part of learning.Partitioning is done based on parameter values, taking one at a time, resulting into equivalent classes with respect to the parameter.During classification, these sets are selected based on parameter values of the test case and the result is collated.The algorithm supports incremental learning at a cost of O(m).The average case complexity of classification is O(m (Iog 2 nf).