How Can a Loss of Information in Mixed Attribute Datasets be Prevented?

Master's Thesis from the year 2012 in the subject Computer Science - General, grade: 1.00, Avinashilingam University, language: English, abstract: This work is concerned with the question of how loss of information in data mining can be prevented by putting in missing values in mixed attributed datasets. Missing value imputation is a procedure that replaces the missing values with some feasible values. Missing data imputation methods are based on only complete instances,instances without missing values in a dataset that is, when estimating plausible values for the missing values in the dataset. Actually, the information within incomplete instances can also play an important role in missing value imputation. Missing data imputation aims at providing estimations for missing values by reasoning from observed data. Because missing values can result in bias that impacts on the quality of learned patterns and the performance of classifications Various techniques have been developed to deal with missing values in data sets with homogenous attributes. But those approaches are independent of all either continuous or discrete value. Moreover these algorithms cannot be applied to real data sets such as equipment maintenance datasets, industrial data sets and gene datasets due to the fact that these data sets contain both discrete and continuous attributes. In order to overcome the above shortcomings, imputation is done in the following manner in this work, there by contributing to both continuous and discrete data. In this method two consistent estimators for discrete and continuous missing target values are developed, and then a spherical kernel based iterative estimator using spherical kernel with RBF kernel and spherical kernel with poly kernel is advocated to impute mixed-attribute data sets, thereby improving the interpolation and extrapolation abilities. The performance of this technique is compared by implementing the imputation with the K-NN, Frequency estimator, RBF kernel, Poly kernel and a mixed kernel and is evaluated in terms of RMSE, which reads out as Root mean square error, and correlation coefficient. In these datasets, the missing values are imputed using higher order kernel functions and the performance is evaluated. From the experimental results it has been observed that spherical kernel with rbf and spherical kernel with poly kernel imputes missing values better when compared to other techniques.