IoT Sensor Data Consistency using Deep Learning

Abstract

Sensor data consistency in Internet of Things (IoT) Applications is the problem of ensuring that large number of sensors in a system are providing mutually consistent values. Detection of data inconsistency can be used to detect unusual conditions like malicious intrusion and other anomalous situation. Machine learning-based anomaly detection approaches can be used to detect sensor data inconsistency. This paper studies the problem of sensor data consistency in the context of detecting hotspots in sensor data being generated in pairs of sensors embedded in a commercial IoT system deployed to monitor grain in large horizontal grain bins. The paper explores how well traditional anomaly detection machine learning algorithms like Location Factor, Isolation Forest, and One class support vector machine work in this environment. A memory efficient Long Short-Term Memory (LSTM) deep learning model was proposed that outperformed the traditional machine learning approaches.