Studies have shown that students have different help-seeking behavior patterns and tendencies and furthermore, that students with certain help-seeking behavior patterns and tendencies may have poor performance (i.e., at-risk students). This study applied an educational data mining approach, including clustering and classification, to analyze students' problem-solving and help-seeking data in a computer assisted learning system to identify student help-seeking behavior patterns and tendencies. First, nine observable problem-solving and help-seeking features for identifying help-seeking behavior patterns were established. Second, this study applied the k-means clustering method and identified three well-known help-seeking behavior patterns: executive, avoidant, and instrumental help-seeking. The results further identified two new help-seeking behavior patterns. One was static instrumental help-seeking and the other was static instrumental and executive help-seeking. Third, executive help-seeking and static instrumental and executive help-seeking patterns could be used as at-risk predicators of poor performance. Fourth, the study applied clustered and identified results to build a minimum distance classifier to identify help-seeking behavior patterns in new data. The study also investigated the accuracy of the classifier in early identifying help-seeking behavior patterns from early-stage data. The early identification accuracy was 61% for the first three minutes and 75% for the seven-minutes of early-stage data, respectively. Fifth, this study identified three help-seeking tendencies: independent problem-solvers, executive help-seekers, and static instrumental and executive help-seekers. In summary, the study showed the feasibility and effectiveness of applying an educational data mining approach, including clustering and classification, to build data-driven student models to identify student help-seeking behavior patterns and tendencies.