Crowdsourcing hazard information to produce valid warnings

Abstract

As major technological advancements are reached, the field of Crowdsourcing keeps being re-discovered in new and innovative applications by researchers and individuals alike. The wisdom of crowds can be harnessed and is a strong tool at a time when social motivational factors, amongst others, are fuelling this rapid development. This project assesses the feasibility of crowdsourcing natural and infrastructural hazard data through a field experiment where recruited participants exploit smartphone environment advantages as part of a proof of concept. This system features capturing of hazard reports, classification and clustering of these reports concluded by the generation and mapping of threats. The domain of crowdsourcing has been extended to hazard analysis, facilitated by established machine learning techniques. This convergence of fields describes the problem of hazard identification as non-trivial. Prior to the implementation, related technologies and existing literature have been reviewed and investigated. This was done to gain a thorough understanding of the fields of study featured in this project. Following the comparison of various algorithms, a decision tree classifier has been implemented on a feature set which includes description, reaction and behavioural attributes alongside a k-means clustering algorithm for the user location coordinates. The final proof of concept has produced decision tree accuracy rates of 94.63% and a precision of 0.82 for the k-means algorithm. Clustering algorithms without a pre-set cluster value failed to predict the true positive cluster amount. Statistically significant conclusions have been drawn from reports collected via a mobile web application. Demographic analysis has been conducted and the usability of the mobile web application has been assessed and achieved a system usability score of 80.48. Such work provides a foundation for future research where the convergence of several fields of study featuring crowdsourcing can contribute to the field of citizen science, specifically hazard analysis.