Abstract :
The incomplete data in which certain features are missing from particular feature vectors, exists in a wide range of fields, including clustering, computer vision, biological systems, classification systems, remote sensing and so on. Classification is a very important research topic in machine learning. There are many algorithms for classification of the numerical data. But the existence of incomplete data degrades the learning quality of classification models and the incomplete data is very common in real world. Usually, classification problem can be separated into two phases: learning phase and classification phase. Many methods dealing with incomplete data in classification problem have been proposed, but most of them only focus on the processing of handling incomplete data in the learning phases. For the incomplete value appearing in the classification phases, almost all of the current approaches can not work. So handling incomplete data in both learning phase and classification phase is important and necessary to be applied for solving the real world problems. In this paper some functions derived from the optimal completion strategy (OCS) is used to estimate the incomplete data. (R.J. Hathaway and J.C. Bezdek, 2001) Then a new method is proposed to classify the incomplete data and it has an outstanding performance. At the same time, this new method can solve two other important problems: rule refinement problem and importance preference problem. Significantly, this is a very powerful and important classifier which can solve all these problems at the same time so well.
