Classification of Incomplete Data by Observation

There are occasions in which databases have feature values that are missing due to errors, irregularities, or unavailable data. Most current imputation methods address cases in which there are sufficient known data to infer an estimate of the missing feature data. This paper proposes a novel imputation algorithm that provides a reasonable solution for the problem domain, represented by databases with missing numerical feature values. This method derives imputed values by observing system configurations and parameters. Given an appropriate model, databases may also be observed. The proposed algorithm employs a weightless multi-classifier that is designed to process certain benchmark databases. Finally, the experiments demonstrate that databases with missing feature values and imbalanced data distributions can still be used effectively.