Which CS1 Students Will Fail? Identifying Digital Markers from Learning Analytics in Computer Systems and Architecture Using Weighted Academic Momentum and Interaction Logs

In the vast landscape of modern education, a quiet revolution is taking place within the digital systems that manage our classrooms. For decades, teachers have relied on grades and attendance sheets to gauge how well a student is doing, but these traditional measures often arrive too late to help a struggling learner. Today, as universities increasingly move their courses online, every click, every login, and every moment spent on a digital platform leaves a trace. Researchers call these traces "digital markers." They are the footprints of a student's daily life in a course, offering a glimpse into their habits, engagement, and potential struggles long before a final exam is ever written. The question driving this new field of study is simple yet profound: can we read these digital footprints to spot a student who is about to fail, and can we do it early enough to change their path?

This is the challenge faced by a team of researchers at the University of Zambia, who turned their attention to a notoriously difficult first-year course: Computer Systems and Architecture. In this class, nearly forty percent of students historically do not pass, a rate that leaves many bright minds behind before their studies have truly begun. The instructors knew that something was wrong, but they lacked a way to identify who was at risk until it was too late. To solve this, the team decided to look beyond the final grades and instead examine the digital breadcrumbs left behind by students over four years. They gathered data from nearly three hundred students, combining their basic background information, their scores on early quizzes and tests, and the raw logs of how they interacted with the university's online learning system. Their goal was to build a model that could act as an early warning system, flagging students who were likely to fail so that teachers could step in with support while there was still time to turn things around.

The researchers began by listening to the people who knew the course best: the instructors, the tutors, and the students themselves. Through interviews and surveys, they identified ten factors that seemed to influence success, such as time management, motivation, and whether a student owned a computer. They then translated these human insights into data points, creating a comprehensive list of features to test. They engineered a special score called "weighted academic momentum," which looked at how a student performed on their very first quizzes and tests, giving more importance to the later, more comprehensive early assessments. They also tracked whether a student had ever logged into the online system at all, and how many distinct days they had been active. By feeding all this information into a computer model, they tested which combination of factors could best predict who would fail.

What they discovered was both surprising and reassuring. The most powerful predictor of failure was not a complex survey about a student's motivation or their prior experience with computers, but rather a simple combination of three things: their early academic momentum, their basic background, and a single yes-or-no question about whether they had ever touched the online learning system. The model found that a student's performance on the first few weeks of assessments was the strongest signal of their future success. However, this signal was even more telling when combined with a simple check of their digital presence. If a student had high early scores but had never logged into the online system, they were still at significant risk. This suggested that some students might be coasting on natural ability or prior knowledge, unaware that they were missing the support structures available online, or perhaps overconfident in their ability to pass without engaging with the course materials.

When the researchers tested their final model on a group of students it had never seen before, it proved remarkably effective at its primary job: finding the students who were in trouble. The system correctly identified eighty-seven percent of the students who eventually failed the course. This high success rate came with a trade-off: the model also flagged some students who would have passed as being at risk, resulting in a false alarm rate of forty-one percent. The researchers argue that this is an acceptable price to pay in an educational setting. It is far better to offer extra help to a student who does not need it than to miss a student who is silently struggling and about to fail. The model achieved this by looking at the data available just five weeks into the semester, a timeframe that gives instructors a crucial window to intervene.

The study also revealed that adding more complex data did not necessarily make the prediction better. Including detailed survey responses about a student's feelings or their specific course workload did not improve the model's accuracy. In fact, the simplest approach—focusing on early grades and a basic check of online activity—worked best. This suggests that for this specific course, the actions a student takes in the first few weeks are a more reliable indicator of their future than their self-reported feelings or background details. The researchers used a method called SHAP analysis to understand why the model made its decisions, which confirmed that the weighted academic momentum was the single most important factor, followed closely by the interaction between that momentum and whether the student had engaged with the online system.

This work offers a practical blueprint for how universities can use the data they already have to support their students. By focusing on simple, early digital markers, educators can move from reacting to failure to preventing it. The researchers plan to turn their findings into a live dashboard that will automatically alert instructors when a student shows signs of trouble, providing them with clear reasons for the warning so they can offer targeted help. While the study was conducted in a single university in Zambia, the approach suggests a universal truth: in the digital age, the way a student shows up, even in the smallest ways, often speaks louder than their background or their initial confidence. The path to success is not always hidden in complex statistics, but sometimes in the quiet, consistent rhythm of showing up to class, both in person and online.