Missing Data Imputation Approach for IoT Using Machine Learning

Over the last decade, numerous cutting-edge technologies have emerged, triggering unprecedented progress in both physical and cyber systems. Among these technologies, connected devices are increasingly integrated into various environments for monitoring and data collection purposes. Missing data imputation remains one of the main challenges in applications utilizing these devices, leading to imprecision of the derived insights and inefficiency of related systems. Recently, numerous machine learning methods and techniques have been proposed to address this challenge. However, these solutions often rely on the use of historical data or the use of records of other features that are highly correlated with the feature containing the missing value. These approaches often overlook large subsequences of missing values as well as missing values in correlated features. This study proposes a new approach to address the missing data issue, assuming that data are collected in several locations. The approach utilizes historical values of the corresponding feature from other locations where data have been recorded with no missing values. One of the aims of this study is to conduct the data imputation process at the edge level, considering the limitations in processing capabilities of edge devices. Several popular lightweight machine learning and deep learning algorithms have been compared in terms of inference time and accuracy.