A First Approach for Handling Uncertainty in Citizen Science

Citizen Science is coming to the forefront of scientific research as a valuable method for large-scale processing of data. New technologies in fields such as astronomy or bio-sciences generate tons of data, for which a thorough expert analysis is no longer feasible. In contrast, communities of volunteers coordinated by the Internet are showing a great potential in completing such analysis in a reasonable time. However, this approach brings uncertainty and the spread of biases within the data, since amateur participants are usually non-experts on the subject and count with variable skills and expertise. This means lack of accuracy in results coming from Citizen Science projects. This work presents a novel approach to handle uncertainty in Citizen Science. We focus on leveraging this uncertainty in the data pursuing a refinement of results. We distinguish between two types of uncertainty: a first one due to the lack of consensus between amateurs, and another one quantified by amateurs themselves during the course of the project. We test our method using the Galaxy Zoo, a project which aims for the labelling of a huge dataset of galaxy images. Considering available expert classifications to validate our experiments, the proposed method is able to improve current accuracy and classify a greater number of images.