loading page

Comparing supervised machine learning approaches to automatically code learning designs in mobile learning
  • +1
  • Gerti Pishtari,
  • Luis P. Prieto,
  • María Jesús Rodríguez-Triana,
  • Roberto Martinez-Maldonado
Gerti Pishtari
Tallinn University

Corresponding Author:gertipishtari@gmail.com

Author Profile
Luis P. Prieto
Tallinn University
María Jesús Rodríguez-Triana
Tallinn University
Roberto Martinez-Maldonado
Monash University


To understand and support teachers' design practices, researchers in Learning Design manually analyse small sets of design ar-tifacts produced by teachers. This demands substantial manual work and provides a narrow view of the community of teachers behind the designs. This paper compares the performance of different Supervised Machine Learning (SML) approaches to automatically code datasets of learning designs. For this purpose, we extracted a subset of learning designs (i.e., their textual content) from Avastusrada and Smartzoos, two mobile learning tools. Later, we manually coded it guided by relevant theoretical models to the context of mobile learning and used it to train and compare several combinations of SML models and feature extraction techniques. Results show that such models can reliably code learning design datasets and could be used to understand the learning design practices of large communities of teachers in mobile learning and beyond. 1 Mobile Learning from a Learning Design perspective Mobile Learning (m-learning) activities promote authentic and contextualized learning [18, 12]. These activities usually take place across spaces (physical and digital) and settings (formal, informal, or non-formal) [9, 11]. To enable teachers to design for m-learning, the field of Learning Design (LD) has come up with several authoring tools [13]. For instance, Smartzoos support the design of geo-localised learning activities outdoor [14], while with GLUESP-AR teachers design activities that happen across multiple physical and digital spaces [9]. Designing learning activities is already a strenuous task for teachers. In m-learning they also have to deal with the complexity of designing across settings and spaces (previously discussed), together with the need to possess substantial technical and pedagogical competencies, relevant to this context. Mettis