The success of data mining relies on the purity of the data set. Before performing the data mining, th e data has to be cleaned. An unprocessed data set may contain noisy or missing values which is a critical researc h issue in the pre-processing stage. Imputation methods are be ing used to solve the missing value problems. In th is proposed work, a machine learning based imputation method is proposed by using the mutual information by exclusively interpolating two different section of the same dataset. For designing the proposed model, a radial basis function based neural network has been used. The performance of the proposed algorithm has been measured with respect to different rate or percenta ge of missing values in the data set and the result s has been compared with existing simple and efficient imputat ion methods also. To evaluate the performance, the standard WDBC data set has been used. The proposed algorithm performs well and was able to impute the missing values even in the worst cases with more th an 50% of missing values. Instead of using simple q uality measure such as Mean Square Error (MSE) to evaluate the imputed data quality, in this study, the quali ty is measured in terms of classification performance. Th e results arrived were more significant and compara ble.
[1]
Dimitris Kanellopoulos,et al.
Data Preprocessing for Supervised Leaning
,
2007
.
[2]
T. Stijnen,et al.
Review: a gentle introduction to imputation of missing values.
,
2006,
Journal of clinical epidemiology.
[3]
James C. Bezdek,et al.
Pattern Recognition with Fuzzy Objective Function Algorithms
,
1981,
Advanced Applications in Pattern Recognition.
[4]
Chengqi Zhang,et al.
Guest Editors' Introduction: Information Enhancement for Data Mining
,
2004,
IEEE Intell. Syst..
[5]
Nicole A. Lazar,et al.
Statistical Analysis With Missing Data
,
2003,
Technometrics.
[6]
Steven F. Messner,et al.
Exploring the consequences of erratic data reporting for cross-national research on homicide
,
1992
.