The ALM project, run by e-learning, pedagogy of mathematics and health care professionals of Aalto University and Arcada University of Applied Sciences, aims to develop an intelligent learning environment for teaching medication calculation. Basic skills of the health care personnel include proficiency in dosage calculation. The challenge of dosage calculation is that the outcome resulting from even a minor error in basic arithmetic may be devastating and pose a serious threat to patient safety. In order to respond to this challenge, the ALM project develops a learning environment for healthcare students and professionals to practise and maintain dosage calculation skills.

The learning environment aims to provide automated, randomized and personalized exercises for students to practise medication calculations. The environment collects data from the students’ performance and uses this information for controlling how the system responds to the each user’s learning needs. The learning environment can adapt to the learning curve of each individual student, and provide the student with exercises most beneficial for his progression in learning.

The development is done by investigating the nature of the medication calculation exercises, and evaluating their usefulness for automated error analysis and data sources for the learning analytics. First the characteristics of the exercises have been studied by means of traditional manual error analysis. The research question is whether the exercises are suitable for automatic analysis performed by the learning environment.

The pedagogical background for the learning environment lies in the 4 Cs teaching model by Johnson et al. In this model, the mathematical proficiency of medication calculation is divided into four areas of competence: computing skills, ability to make unit conversions, understanding to conceptualise the problem, and being able to critically evaluate the outcome. These have been used as the base for categorizing the students’ errors.

For achieving a desirable level of automation, further development is needed. The exercises must be composed so that the outcome of the error analysis indicates students’ cognitive processes. Also, they must support the students’ positive confidence on learning and motivation to study. The interface between the learning environment and the learning analysis tools (used by teachers) must enable easy handling of the error data. The analyzer tool must provide coherent classification and subsequent analysis. This is done by adjusting the classification classes and logic to interpret the answers of the student in a sensible way.

Also, the analyzer must be able to profile the students according to their performance. This allows the learning environment to adjust its operations logics to benefit each student individually. Furthermore, exercises developed with the novel stateful question type that extends STACK with state and inner-loop adaptation allowing a more expressive and game-like series of exercises. Finally, all this data should be stored cumulatively enabling analysis in larger scale, e.g. by means of big data and AI, to better understand the cornerstones of students’ cognitive processes in learning skills of medication calculation.