Abstract
Wireless sensor networks consist of a large number of heterogeneous devices that communicate to collaboratively perform various tasks for users. Heterogeneous devices are deployed to sense the context of the environment. The context information is use to actuate various devices or services to support various activities of a user in a smart environment. Therefore, data correction is vital in managing issues arising from missing or corrupt contextual data due to system internal and external influences. We would like to investigate the machine learning techniques to ensure a complete and accurate sensor dataset for smart environment applications by runtime correcting missing or corrupt data due to sensor failures. We proposed a framework to correct dynamically sensory data. Specifically, we deal with the problems of faulty data (outliers, spikes, stuck-at, and noise), and missing information. Our proposed framework is able to learn temporal correlations in collected data from smart objects using Artificial Neural Network algorithm. We utilize the learned correlations to discover faulty data patterns to recover them, and imitate missing information. We implement the proposed data correction framework and test it on two real-world datasets collected from transportation domain (parking system, and road traffic).