Imputation of outliers and missing values for activated sludge dissolved oxygen database using multivariate imputation by chained equations (mice)

Activated sludge process (ASP) is the most widely used process in wastewater treatment plants. The concentration of dissolved oxygen (DO) is a crucial controlling parameter in this process as the DO affects the efficiency of the treatment, operational cost and system sustainability. While Field electrodes measure DO, maintaining regular and accurate records remains a challenge, even in the absence of aggressive climate conditions. Most of the DO probes are sensitive and require careful maintenance and calibration that may lead to non-accurate data recording. This paper discussed the validity of statistical models to impute missing DO values. Multivariate Imputation by Chained Equations (MICE) is used in R programming language to predict missing values and replace outliers in the DO measurements. The performance of this technique was very good as the correlation ranges between 0.999 and 0.821 when the missing values range from 1 to 15 % (424 to 6,360 of total data of 42,420). The method is highly effective and ready to be used for many other applications. Kohonen Self Organising Map (KSOM) is applied as an alternative model to verify the accuracy of the MICE.