<p>The COVID-19 pandemic has accelerated the shift toward online education, emphasizing the critical role of Learning Management Systems (LMS) and driving extensive research in Learning Analytics. LMS platforms automatically capture large-scale learner interaction data, facilitating advanced analyses through statistical methods and artificial intelligence techniques. However, simplistic interpretations of LMS log data can neglect contextual nuances, potentially leading to misleading results. This study examines an advanced AI education program characterized by minimal instructor intervention, requiring learners to depend heavily on Self-Regulated Learning and effective learning strategies for successful course completion within limited timeframes. Specifically, the research identifies methodological pitfalls inherent in LMS log data analyses and proposes improvements through comprehensive clustering of learning behavior patterns and predictive modeling using early-stage LMS data. Findings reveal that distinct learning strategies significantly affect course completion outcomes, while predictive models employing early interaction data effectively identify learners at high risk of course non-completion. Furthermore, Shapley Additive Explanations highlight critical intervals in learning trajectories, pinpointing key lecture segments where targeted interventions could enhance learner retention. This study advances Learning Analytics theoretically by refining analytical methodologies and practically by enabling data-driven instructional design and personalized learner support, underscoring the potential of adaptive strategies in improving course completion outcomes in LMS-based AI education.</p>