Proactive missing values imputation based on reinforcement learning

Abstract The huge growth and use of the Internet of Things (IoT) in everyday activities in combination with the limited resources present in the involved devices has induced the adoption of Edge Computing (EC) that offers a vast infrastructure where the data can be collected and processed before they are transferred to the Cloud. In the EC infrastructure, there is an increased number of nodes that enhance the pervasiveness of services and applications as EC nodes are present close to end users. From a global view, the IoT infrastructure plays the role of the data collector while the EC nodes act as intermediate processing points where many tasks can be executed. One significant problem is the presence of missing values in the collected data. In this paper, we propose an ensemble-based model for missing values imputation. We adopt a Reinforcement Learning mechanism for the proactive detection of the nodes that have similar behavior to the node that detects missing values. We aim to perform the envisioned imputation value through the aggregation of the suggested replacements based on the local dataset and the weighted replacements as delivered by the group, i.e., nodes with similar data. Moreover, we elaborate on a weighted mechanism for the assignment of proper weights in the nodes with similar datasets as they are depicted by historical records. We evaluate our model by adopting various experimental scenarios.