Learning Analytics (LA) can be utilized to detect real-time indicators of struggle in novice programming students, particularly within remote learning environments. With the increasing use of AI tools among students, challenges have emerged in the field of programming education regarding students’ false perceptions of understanding and reliance on AI-generated solutions. This study explores how selected indicators, including time on task, task completion, paste behaviour, and keystroke patterns, as well as AI usage correlate with novice programming students' struggles during task completion. Additionally investigating how do students’ self-perception of their programming skills compare to their actual knowledge. To answer these questions a mixed-method approach was designed involving behavioural data collection via a coding platform PixieCo/de which was modified for the purposes of this study. The method combines data from a self-assessment survey, a programming knowledge quiz, task execution and post-task interviews. 

The results suggest that keystroke data, when analysed through sequence mining and triangulated with other indicators, was the most reliable indicator to reflect levels of student engagement and difficulty and offered the most nuanced insights. These findings highlight the potential of LA to provide timely insights that can support educators and learners alike.  Self-assessment data revealed that most students either accurately or underestimated their abilities, with no observed overconfidence. Future research could investigate how these indicators can be used at a larger scale in classrooms and how LA results could be presented to students to aid their learning processes.