Skeptical Learning—An Algorithm and a Platform for Dealing with Mislabeling in Personal Context Recognition

Mobile Crowd Sensing (MCS) is a novel IoT paradigm where sensor data, as collected by the user’s mobile devices, are integrated with user-generated content, e.g., annotations, self-reports, or images. While providing many advantages, human involvement also brings big challenges, most critical possibly poor quality of human-provided often due to inaccurate input from non-expert users. In this paper, we propose Skeptical Learning, an interactive machine learning algorithm checks user feedback and tries fix it when problem arises. context, consists answers generated questions, at times defined machine. The main idea integrate three core elements, which (i) data , (ii) answers, (iii) existing prior knowledge world, enable second round validation any time these types information jointly generate inconsistency. proposed solution evaluated in project focusing on university student life scenario. goal recognize locations transportation modes students. results highlight unexpectedly high pervasiveness mistakes students project. shows advantages provided Learning dealing mislabeling issues way improving prediction performance.