Issues on critical objects in mining algorithms

Abstract

Data objects are considered as fundamental keys in learning methods that without the objects the mining algorithms are meaningless. Data objects basically direct the accuracy of the selected algorithm in case if they are extracted from inappropriate groups. Knowing the exact type of data object leads the miner to provide a suitable environment for learning algorithms. Supervised and unsupervised learning methods propose some membership functions that perform with respect to behaviour of each data category to classify data objects and solutions. The paper explores different type of data objects by categorizing them based on their behaviour with respect to learning methods. We also introduce some critical objects that play the main role in each data set. Issues on critical objects in mining algorithms are fully discussed in this paper. The accuracy and behaviour of these critical objects are compared by running fuzzy, probabilistic, and possibilistic algorithms on some data sets presented in this paper. The results prove that some methods are able to provide a suitable environment for critical objects and some are not. The comparison results also show that most of the learning methods have difficulties dealing with critical objects. Lack of ability to deal with these objects may cause irreparable consequences.