ITCI:An information theory based classification algorithm for incomplete data

In the field of data mining, classification is an important aspect which has been studied widely. However, most of the existing studies assumed the data for classification is complete, while in practice, a lot of data with missing values exists. When dealing with these data, deleting the incomplete instances will result in a reduction of available information and filling in missing values may introduce skew and errors. To avoid the above problems, it is of great importance to study how to classify directly with incomplete data. In the paper, an information theory based classification algorithm, ITCI, is proposed. ITCI calculates the initial uncertainty of each class and attributes' contribution to decrease class uncertainty in the training stage and then, in the testing stage, an instance is assigned to the class whose uncertainty is minimum after all of the attributes are taken into consideration. Extended experiments proved the effectiveness and feasibility of the proposed method.