Learned helplessness (LH) in mathematics can hinder student performance and engagement, particularly in digital learning environments. This study aimed to identify both adaptive and maladaptive learning behaviors and provide insights to inform interventions that foster persistence and self-regulated learning. It also contributes to research on human–computer interaction in underrepresented regions, specifically the Philippines, where mathematics achievement remains among the lowest in international assessments. Using Latent Profile Analysis (LPA), the study identified eight distinct student engagement profiles based on interactions with a mobile mathematics learning app, each varying in problem-solving persistence, hint reliance, and disengagement patterns. Statistical analyses confirmed significant behavioral differences across profiles, although high and low LH students differed only in time spent, and this difference did not remain significant after correction. Prediction models showed that LPA profiles significantly predicted problem-skipping and mistake occurrence, while hint usage and time spent were less reliably associated with profile classification. These findings suggest that LH may manifest through inefficient strategies rather than complete disengagement and highlights the need for targeted interventions that promote productive engagement in mobile mathematics learning.