A Semi-supervised Learning Imputation Model for Automatic Identification System Data

Abstract

The quality of Automatic Identification System (AIS) data is weakened as a result of geographical, environmental, and technical constraints. The presence of these constraints hinders the reliable collection of AIS data, leading to a decline in data quality, particularly evident through the occurrence of missing values. Missing values in AIS data occur sporadically and pose a challenge when it comes to replacing them. However, the frequently employed approach of Supervised Learning model has limitations in dealing with such irregular occurrences. Therefore, in this study, an imputation method utilizing Semi-Supervised Learning Imputation (SSLI) model has been introduced. Additionally, a novel loss function has been proposed for SSLI. The method proposed in this paper shows better performance in replacing missing values, especially in situations when the pattern is complicated, by using information from both previous and subsequent periods during the learning process. The imputation model proposed in this study has the potential to improve the quality of AIS data for ships with complex patterns. This expected benefit extends to other research applications, enhancing its utility across a broad spectrum of disciplines.