A Method of Pruning and Random Replacing of Known Values for Comparing Missing Data Imputation Models for Incomplete Air Quality Time Series

The data obtained from air quality monitoring stations, which are used to carry out studies using mining techniques, present the problem of missing values. This paper describes a research work on imputation. Among most common methods, method that best imputes values available set is analysed. It uses an algorithm randomly replaces all known in dataset once with imputed and compares them actual values, forming several subsets. Data seven stations Silesian region (Poland) were analyzed for hourly concentrations four pollutants: nitrogen dioxide (NO2), oxides (NOx), particles 10 μm or less (PM10) sulphur (SO2) five years. Imputations performed linear imputation (LI), predictive mean matching (PMM), random forest (RF), k-nearest neighbours (k-NN) by Kalman smoothing structural time series (Kalman) methods performance evaluations performed. Once comparison was validated, it determine that, general, fitted pattern. observed each behaves analogous way different variables results NO2 SO2. UMI worst imputer sets.