Focalize K-NN: an imputation algorithm for time series datasets

• Published in: Pattern analysis and applications : PAA Vol. 27; no. 2
• Main Authors: Almeida, Ana, Brás, Susana, Sargento, Susana, Pinto, Filipe Cabral
• Format: Journal Article
• Language: English
• Published: London Springer London 01.06.2024; Springer Nature B.V
• Subjects: Algorithms; Computer Science; Datasets; Decision making; Missing data; Original Article; Pattern Recognition; Time series; Missing data imputation; K-nearest neighbors; Machine learning; Time series
• ISSN: 1433-7541; 1433-755X
• DOI: 10.1007/s10044-024-01262-3

The effective use of time series data is crucial in business decision-making. Temporal data reveals temporal trends and patterns, enabling decision-makers to make informed decisions and prevent potential problems. However, missing values in time series data can interfere with the analysis and lead to inaccurate conclusions. Thus, our work proposes a Focalize K-NN method that leverages time series properties to perform missing data imputation. This approach shows the benefits of taking advantage of correlated features and temporal lags to improve the performance of the traditional K-NN imputer. A similar approach could be employed in other methods. We tested this approach with two datasets, various parameter and feature combinations, and observed that it is beneficial in scenarios with disjoint missing patterns. Our findings demonstrate the effectiveness of Focalize K-NN for imputing missing values in time series data. The more noticeable benefits of our methods occur when there is a high percentage of missing data. However, as the amount of missing data increases, so does the error.