A Novel Framework for Missing Data Imputation and Feature Selection in Incomplete Information Systems Using Hybrid Rough Set Theory

Abstract

Today’s real-world data is highly susceptible to noisy and missing data due to their normally huge size, besides their connection to heterogeneous sources. Decision making can be improved if data is preprocessed. Data cleaning has remarkable significance in data preprocessing. Missing values mostly occurs in any type of data sources. Imputation of missing values with appropriate values is a challenging task. Wrong imputations may affect the decisions observed. The main objective of our article is to investigate the impact of feature selection upon imputation of missing values across various datasets. We have used five distinct datasets with a range of feature sizes while performing experimental tasks. Also, we have compared three distinct forms of feature selection techniques and imputation approaches. The findings demonstrate the effectiveness of an approach that combines feature selection with imputation that may be suitable for a number of medical datasets. In fact, our experimental results show that it is possible to generate best results in case of well- designed feature selection algorithms.