Files
Abstract
Crowdsensing temperature data has enabled a paradigm shift in the ways we collect data and analyze the heat exposure effects on individuals and small communities. Use of low-cost sensors has helped in gathering granular spatial-temporal temperature data and capturing ever-changing ambient environmental conditions. However, the practice poses challenges such as data integrity, and sensor failures. One of the main concerns is placement of temperature sensors such that they are shielded from the natural environment (for example, in air-conditioned vehicle, inside a bag) during data collection. This will lead to anomalous data collection. We propose a novel approach to detect anomalous sensor placement based on empirical observations, temperature readings of a sensor exposed to the natural environment show more frequent fluctuations than temperature readings of a sensor shielded from it. We use sliding window technique and supervised learning classifier to detect anomalous temporal temperature subsequences effectively. We also do comparative performance analysis of SVM, Logistic Regression and Random Forest classifiers.