An ensemble of rule-based classifiers for incomplete data

Many real-world datasets suffer from the problem of missing values. Imputation which replaces missing values with plausible values is a major method for classification with data containing missing values. However, powerful imputation methods including multiple imputation are usually computationally intensive for estimating missing values in unseen incomplete instances. Rule-based classification algorithms have been widely used in data mining, but the majority of them are not able to directly work with data containing missing values. This paper proposes an approach to effectively combining multiple imputation, feature selection and rule-based classification to construct a set of classifiers, which can be used to classify any incomplete instance without requiring imputation. Empirical results show that the method not only can be more accurate than other common methods, but can also be faster to classify new instances than the other methods.