Missing Value Imputation Approach Using Cosine Similarity Measure

Mining incomplete datasets containing missing values can produce various problems in data mining, particularly with the large-scale dataset. Missing data is one of the major factors affecting the quality of data. Missing value imputation provides optimal solutions to deal with incomplete datasets containing missing values. Although the class center-based approach for imputing missing value provides good results, however, they use Euclidean distance which works on linear distances which generate the linearity in distance: hence making the missing data imputation less accurate. In this paper, missing value imputation approach using cosine similarity measure is used. The proposed model uses the implicit classification label estimation and cosine similarity ratio formulation which provides better results than class center-based imputation approach.