Digital learning platforms are promising large scale experimental fields to design well controlled experiments. They also provide large databases for researchers in psychology and educational data mining.This project aimed at : i) shedding light on underlying behavioral learner patterns; ii) improving learner predictive modeling to better tailor adaptive online tutoring systems. Learner modeling techniques are crucial for providing a personalized and more efficient adaptive instruction to learners, especially on Intelligent Tutoring Systems (ITS).In this study, we analyzed data from an experiment investigating the combined effect of the restudying strategy (reading vs. testing) and course grain size on learner memory retention in two delayed tests (7 and 28 days after initial learning).  The experiment took place on the e-learning platform “Didask” (https://www.didask.com/) and the learning content concerned cognitive balance at work. More than 300 learners participated in the study. After aggregation, four types of variables were considered in relation to the final performances: i) behavioral (e.g. learning durations), ii) socio-demographic (e.g. age, educational background),  iii) metacognitive abilities (Metacognitive Awareness Inventory form), iv) motivational (i.e., being interested in the experiment results). For instance, we performed a hierarchical cluster analysis (HCA) on the learners and identified several homogeneous learner profiles. We finally investigated the generalization of our findings by comparing them with a previous experiment involving similar learning conditions but with less complex information to learn. Data from the second final exam has not yet been collected, but we expect it to shed light on deeper learner patterns.