Deep learning for automatic usability evaluations based on images: A case study of the usability heuristics of thermostats Academic Article in Scopus uri icon

abstract

  • © 2017 Elsevier B.V.Thermostats are designed for increasing requirements on indoor thermal comfort. Nevertheless, they are critical devices for saving energy in buildings and households. However, when thermostats do not accomplish the usability requirements, the end-users do not save energy. Then, when a thermostat is designed or validated, one of the leading problems that must be tackled is the usability evaluation. Generally, the evaluation is based on usability heuristics that are done by experts and designers and involve a very complicated cycling process in which usability experts need to be included in the complete usability evaluation. On the other hand, there are several proposals for generating an automatic usability analysis that can be used by designers or end-users. However, they are limited by the methodologies that are implemented in the evaluation because usability evaluations necessitate a large amount of data abstraction, and the amount of processed information is enormous; As an alternative, Artificial Intelligence can help to solve this problem, especially machine learning techniques with deep learning capabilities that can reach a high level of data abstraction with a significant amount of information and implement an automatic usability evaluation based on images. Convolutional networks that are included in deep learning can classify complex problems, attain highly accurate results. This paper proposes to train a convolutional network with standard usability heuristics for evaluating usability, which is an easy method for evaluating usability in thermostats, based on images. The proposed automatic method gives excellent results for evaluating usability heuristics in the heuristic assigned. This paper provides a complete methodology, using deep learning, for automatically evaluating the usability heuristics of thermostats.

publication date

  • March 15, 2018