Improving accuracy rate of imputation of missing data using classifier methods

Abstract

Managing missing data is a decisive work to ensure good results in mining. In order to get the complete knowledge of dataset, the imputation technique is required to fill the missing data. A measure has been taken to improve the accuracy of the imputation by considering new imputation method with four other existing methods with six existing classifiers for various amount of missing values ranging from 5 to 55%. This paper explores an idea of how the different imputation method influences the performance of classifiers that are subsequently used with the imputed data. This experiment focuses on discrete data. So as to improve the quality of imputation(1.545% reduced the classification error), Few well known classifiers LSVM, RIPPER, C4.5, SVMR, SVMP, and KNN have been utilized to improve the imputation accuracy. The results shown in this paper confirms that the MMSD imputation method is better among all other methods, it produces the better performance with the classifier upto 7.72% at 20% missing values.