Evaluating Time Series Imputation Algorithms: Kalman Smoothing vs Seasonal Trend Decomposition Using LOESS on Climate Data

Missing data refers to a condition where part of the dataset is unavailable due to unrecorded information during the data collection process. Missing data problems must be addressed because they can affect the accuracy of the analysis. This study aims to evaluate the performance of univariate imputation methods, namely Kalman Smoothing and STL Decomposition, on time series data obtained from a weather station in Lampung during the period 2001–2024. This study also incorporates the application of missing data mechanisms, specifically MCAR (Missing Completely at Random), MAR (Missing at Random), and MNAR (Missing Not at Random), with missing rates of 5%, 10%, 20%, 30%, 40%, and 50%. The variables used in this analysis include average temperature, relative humidity, total rainfall, solar radiation, and wind speed. The results show that STL Decomposition is more effective and accurate than Kalman Smoothing, indicated by lower Root Mean Square Error values. This method performs better at missing rates of 5%, 10%, and 20%, and demonstrates superior performance in handling MCAR, MAR, and MNAR missing data conditions. Although Kalman Smoothing produces stable results, STL Decomposition provides more precise estimates for missing data.