On the Issue of Incomplete and Missing Water-Quality Data in Mine Site Databases: Comparing Three Imputation Methods

Abstract

Large water-quality databases are valuable for predicting mine drainage chemistry, identifying optimal measures for mitigation and remediation, and refuting/refining models and theories. However, such databases often have missing values due to periodic lack of sampling and analysis or input errors. These missing values lead to problems in machine learning and statistical analysis of water-quality data from mine sites. Using water-quality data collected from 1971 to 1994 from many locations at a copper-molybdenum-gold-silver-rhenium mine site, we compared three imputation methods to estimate missing water-quality data: iterative robust model-based imputation (IRMI), multiple imputations of incomplete multivariate data (AMELIA), and sequential imputation for missing values (IMPSEQ). These methods were evaluated based on mean absolute error, relative absolute error, and percent bias techniques. The results showed that IMPSEQ and IRMI are suitable to impute missing values in water-quality databases at mine sites, whereas AMELIA is not.