Multivariate real-time Missing Value Imputation using Adaptive VAR with IoT data from a Refrigerated Container

The sensors used in IoT devices in a cold chain generates massive amount of streaming data such as temperature, humidity and carbon dioxide. This multivariate data often contains missing values dues to communication failure, node malfunctioning, and power fluctuations. Thus, there is a need for faster preprocessing algorithms to deal with this problem for reliable online decision making. This paper presents an adaptive Vector Auto Regressive (VAR) algorithm using recursive least squares for missing data imputation in real-time application. The algorithm is compared with few existing algorithms using the data collected from the sensors installed in a reefer container. The results show adaptive VAR outperforms others in terms of RMSE and correlation as the evaluation metrics. The paper also demonstrates the implementation of adaptive VAR in the edge device.