With the growth of technologies and the line between the traditional and virtual classrooms blurring, learning can take place basically anywhere. Self-Initiated Learning Scenarios are environments that enable students to learn on their own without the supervision of a teacher. Self-regulated learners are students who can greatly-benefit from these environments and as such, in this research their activities are tracked to be able to generate a model for positive learning habits. With the use of an annotation tool called Sidekick, these self-regulated learners undergo a process referred to as self-reflection. This is a phase in the learning process that enables students to reflect and improve on their learning habits.

Twenty five (25) undergraduate computing students participated in the study immersing themselves in these scenarios. These students were assessed and categorized into three levels of self-regulation namely low, moderate and high self-regulation.

A model is created based on their interaction data following a machine learning task. A general model covering three types was developed but only performed with a 48% accuracy leading to the need to develop a class-specific model covering these types of learners. The class models specific for students of low, moderate and high regulation obtained accuracy scores of 63.3%, 55.5% and 50.8% respectively; outperforming significantly better than that of the general model.

Their interaction data and models, along with their transitions were used to generate a set of rules called policies, employing a profit-sharing algorithm. Three sets of policies were generated depending on the type of self-regulated learner the student falls into. These sets of policies agree with the specific class precision and recall values from the models, thereby creating a set of rules for activities that when followed would improve or maintain the motivation of students regardless of self-regulation type that they fall to.

Keywords: SideKick, self-regulated learners, self-initiated learning, annotation, activity recognition.