A Visual Analytics Framework for Imputing Incomplete Time-Series Data

Missing or corrupted data during collection or transmission can distort the original characteristics of time-series datasets, leading to biased and unreliable analyses. Although numerous statistical and machine learning techniques have been developed to address data incompleteness, the absence of ground truth makes it challenging to determine the optimal imputation strategy. The effectiveness of imputation largely depends on the nature of the dataset, the mechanism and extent of missingness, and the specific missing patterns—all of which can vary significantly. Consequently, the quality and reliability of imputation results are often influenced by the analyst’s expertise and intuition. In this paper, we propose a visual analytics approach designed to support decision-making in imputing missing values within incomplete time-series data. Our system enables analysts to investigate potential causes of missingness, understand data characteristics, and compare various imputation models interactively. Through a series of use cases, we demonstrate how our approach facilitates informed and transparent model selection, ultimately enhancing the trustworthiness of downstream analysis.