White-Box Decision Tree Algorithms: A Pilot Study on Perceived Usefulness, Perceived Ease of Use, and Perceived Understanding

Abstract

The mainstream in undergraduate data mining algorithm education is using algorithms as black-boxes with known inputs and outputs, while students have the possibility to adjust parameters. Newly proposed white-box algorithms provide students a deeper insight into the structure of an algorithm, and allow them to assemble algorithms from algorithm design components. In this paper a recently proposed data mining framework for white-box decision tree algorithms design will be evaluated. As the white-box approach has been experimentally proven very useful for producing algorithms that perform better on data, in this paper it is reported how students perceive the white-box approach. An open source data mining platform for white-box algorithm design will be evaluated as technologically enhanced learning tool for teaching decision tree algorithms. An experiment on 51 students was conducted. A repeated measures experiment was done: the students first worked with the black-box approach, and then with the white box approach on the same data mining platform. Student's accuracy and time efficiency were measured. Constructs from the technology acceptance model (TAM) were used to measure the acceptance of the proposed platform. It was concluded that, in comparison to the black-box algorithm approach, there is no difference in perceived usefulness, as well as in the accuracy of produced decision tree models. On the other hand, the black-box approach is easier for users than the white-box approach. However, perceived understanding of white-box algorithms is significantly higher. Evidence is given that the proposed platform could be very useful for student's education in learning data mining algorithms.