A Statistical-Temporal Framework for Evaluating Missing Value Imputation on Daily Waste Data

Time series data with missing values presents a significant challenge in data analysis, particularly in daily waste data, which exhibits unique characteristics such as inconsistent recording patterns, inter-period dependencies, and irregular distributions. This research introduces the Statistical-Temporal Evaluation Framework (STEF), a comprehensive framework designed to evaluate missing value imputation methods. The framework encompasses three main evaluation dimensions: statistical distribution analysis, to ensure the appropriateness of data distribution after imputation; independence and identically distribution (IID) analysis using the Durbin-Watson and Kolmogorov-Smirnov tests, to evaluate independence and identical distribution properties; and Time Lag Analysis (TLA), to assess the temporal relationships between periods using the Autocorrelation Function (ACF). In this study, STEF was applied to daily waste data with single missing values and complex patterns. Five imputation methods were tested: linear interpolation, polynomial interpolation, KNN, Kalman Filter, and polynomial regression. The results identified linear interpolation as the best imputation method, based on the combined metric values and standard error. The evaluation demonstrated that the STEF framework effectively considers both the statistical and temporal characteristics of the data, providing more accurate and representative imputation results that align closely with the actual data.