One-to-one online tutoring provided by human tutors can improve students' learning outcomes. However, monitoring the quality of such tutoring is a significant challenge. In this paper, we propose a learning analytics approach to monitoring online one-to-one tutoring quality. The approach analyzes teacher behaviours and classifies tutoring sessions into those that are effective and those that are not effective. More specifically, we use sequential behaviour pattern mining to analyze tutoring sessions using the CM-SPAM algorithm and classify tutoring sessions into effective and less effective using the J-48 and JRIP decision tree classifiers. To show the feasibility of the approach, we analyzed data from 2,250 minutes of online one-to-one primary math tutoring sessions with 44 tutors from eight schools. The results showed that the approach can classify tutors' effectiveness with high accuracy (F measures of 0.89 and 0.98 were achieved). The results also showed that effective tutors present significantly more frequent hint provision and proactive planning behaviours than their less-effective colleagues in these online one-to-one sessions. Furthermore, effective tutors sequence their monitoring actions with appropriate pauses and initiations of students' self-correction behaviours. We conclude that the proposed approach is feasible for monitoring the quality of online one-to-one primary math tutoring sessions.