We analyzed data from student use of the Learning Management System (LMS) in a large-enrollment undergraduate chemistry course and examined the relationships among social and emotional (SE) skills, as measured by ACT® Tessera®, course features, and course grades. Our goal was to understand how SE skills are associated with students’ online behaviors and course outcomes and to explore a principled design approach for feature extraction using individual activities and sequential data mining techniques. We found low to moderate associations (r = 0.10 - 0.34) between these features and SE skills (Grit, Teamwork, Resilience, Curiosity, and Leadership) and between these features and course grade (r = 0.18 - 0.36). These behavioral features (i.e., individual activities and learning tactics) were used to create predictive models that accurately predicted SE skills (root mean squared error [RMSE] = 0.87) and course grade (RMSE = 0.67). These models had higher predictive accuracies than baseline models created from only student background information (including college grade point average [GPA]). This finding suggests a new source of insights that can be used to create accurate and actionable predictions to improve student course grades and SE skills. Furthermore, a mediation analysis revealed a significant partial mediation effect of student online behaviors on the influence of SE skills in final course grade. This suggests that SE skills may have some influence on the daily behaviors of students, which in turn are related to class outcomes and student success. No bias based on student demographic background was observed in our predictive models.