A decade ago, learning analytics was a niche term confined to research papers. Predictive models for student performance were largely experimental. Today, institutions have the data, the tools, and the AI capability to know which students are struggling.
The situation confronting educators is this: by the time a final exam result is recorded, the window for meaningful intervention has often already closed.
This is no longer just a tracking problem. In every program from licensure-driven courses to general education and technical training, institutions are sitting on data that could flag risk early. Still, most wait for the semester to end before acting. The cost isn’t only academic. It’s lost time, lost confidence, and in many cases, a student who didn’t have to fall behind.
The strengths of traditional assessment — and their limitations
Final exams, term papers, and grade reports continue to matter. They measure mastery, validate readiness, and give institutions a clear academic record. Their value is not in doubt.
What they find difficult to do, however, is warn early enough. A student’s struggles in week three are rarely visible in a system that only reports outcomes in week sixteen. By the time a failing grade appears on record, the opportunity to intervene has already passed.
Today, institutions are less concerned about whether they can measure outcomes. They are asking: can we see the problem while there’s still time to fix it?
AI and learning analytics as an early-warning resource for institutions
AI in this context is not designed to replace instructors or assessments. It is intended to make existing learning data more useful and to turn scattered indicators into a system that flags risk before risk becomes failure.
Learning analytics has been defined as the measurement, collection, analysis, and reporting of data about learners and their contexts to understand and optimize learning environments. It draws from machine learning, artificial intelligence, educational data mining, visualization, social network analysis, e-learning, psychology, and educational theory [1]. In practical terms, institutions can use learner-produced data not only to describe what already happened, but to predict risk, recommend intervention, and personalize support — while a course is still in progress.
When a student’s quiz scores begin slipping, an alert can prompt outreach. When attendance drops before a major exam, an adviser can be notified. When mock board results stagnate across review cycles, a targeted study plan can be triggered. Institutions can respond in days instead of waiting for a semester-end report.
Reasons for urgency in Philippine higher education
The urgency is particularly evident in licensure-driven programs and competitive academic tracks. Student performance is shaped by many interacting factors: prior academic background, course-level performance, internal assessments, attendance, engagement, learning behavior, and consistency over time. Studies on student performance prediction commonly use CGPA, internal assessments, quizzes, tests, attendance, demographic data, external assessment results, and course grades to forecast academic success [2].
The result, without early tracking, is a familiar pattern: students who appear fine on paper until a board exam or capstone reveals otherwise. Research on licensure examination performance has shown that academic averages, mock board examination results, and self-review behavior can predict eventual outcomes — and that prediction systems should never replace face-to-face instruction, but should enhance it by identifying learners who need additional support, earlier [3].
AI-supported tracking provides a functional link — one that connects scattered performance signals into a single, actionable picture for academic advisers and program heads.
Technology as the foundation of a system that actually intervenes
Predictive analytics is only useful at scale when it’s connected to real action. Without that connection, dashboards and risk scores are just numbers nobody acts on. The right system turns data into decisions.
In practice, AI-supported tracking works across several layers. At the first level are performance indicators as grades, assessment scores, mock examination results, subject-level mastery, etc. Second are behavioral indicators like attendance, learning platform activity, time on task, frequency of engagement, etc. Finally are the progression indicators: improvement over time, consistency of performance, and changes in learning behavior.
Together, these layers move institutions past a single question — did the student pass or fail?
Who may be at risk?
Which competencies require support?
Is the learner improving, declining, or remaining stagnant?
Is performance consistent or unstable?
What intervention should be provided now?
A review of learning analytics models shows such systems can support monitoring, feedback, forecasting, assessment, personalization, recommendation, mentoring, guidance, and intervention. These goals directly shape how teachers and academic advisers support learners, not just how institutions report to themselves [4].
Activity is not the same as learning
Institutions adopting AI must guard against a common trap: mistaking activity for progress. Recent research on AI chatbot use in higher education offers a clear warning sign. Students who used AI chatbots completed reasoning tasks faster, but did not produce significantly better written responses than non-users. Regular AI users passed more exams, but did not necessarily earn better grades. AI may improve speed and throughput without improving understanding [5].
This distinction matters. A student who submits faster or logs in more often isn’t automatically learning more deeply. That’s why AI-supported tracking should monitor both quantity indicators — participation, completed assessments, time on task, engagement frequency, completion rates — and quality indicators — grades, competency mastery, reasoning quality, improvement trends, and the ability to apply knowledge in real contexts.
Predictive models can also help institutions identify which data points actually matter. In one supervised data mining study comparing K-Nearest Neighbor, Naïve Bayes, Decision Tree, and Logistic Regression, Naïve Bayes performed best in predicting whether students would be excellent or non-excellent — and specific course grades emerged as among the most important predictors [6].
There is no universal formula. In one program, early quiz performance may be the strongest signal. In another, attendance and engagement carry more weight. In licensure-driven programs, mock examination performance and subject-specific mastery often matter most. Each institution has to find its own signal.
This isn’t about replacing educators with algorithms
Organizations can keep their assessment systems, their grading structures, and their instructors at the center of the classroom. The objective isn’t automation for its own sake — it’s giving educators a clearer, earlier view of what’s actually happening with their students.
A responsible approach to AI in tracking learning outcomes rests on a few principles. Institutions should define learning outcome indicators beyond grades alone — competency mastery, progression, retention, and readiness for professional practice. They should draw from multiple data sources, since a single exam score rarely tells the full story. They should separate learning activity from learning quality, monitoring both participation and mastery. They should connect every predictive insight to a real intervention — advising, review sessions, adaptive materials, mentoring. And they should apply ethical safeguards throughout, so predictions widen access to support rather than permanently labeling students or narrowing their opportunities.
The real question is not “if” — it is “how early”
Risk to student outcomes is no longer hypothetical. It shows up in failed board exams, in stalled grades, in students who finish a term without anyone having noticed the warning signs along the way.
Institutions need to consider not whether AI-supported tracking is worth adopting. It’s a question of how early they can see risk, how reliably their data points to real intervention, and how seamlessly that insight reaches the educators and advisers who can act on it.
As programs move forward, students are counting on their institutions to catch the signal before it becomes a statistic.
Interested in exploring how AI-powered learning analytics can strengthen outcome tracking at your institution? Board Prep Solutions invites school administrators, academic leaders, and educators to start the conversation. Reach out to learn how the right platform can help your school catch risk early — and turn data into timely support for every learner.
Endnotes
- Ranjeeth, S., Latchoumi, T. P., & Victer Paul, P. The article explains that learning analytics involves the measurement, collection, analysis, and reporting of learner and contextual data to understand and optimize learning environments, drawing from machine learning, AI, educational data mining, visualization, social network analysis, e-learning, psychology, and educational theory.
- Yaacob, W. F. W., Nasir, S. A. M., Yaacob, W. F. W., & Sobri, N. M. The article identifies commonly used attributes in predicting student performance, including CGPA, internal assessment, quizzes, tests, attendance, demographic data, external assessment, and course grades.
- Tarun, I. M. The study used data mining classifiers to generate licensure examination performance prediction models, integrated into an online test and decision-support system, finding that academic averages, mock board examination results, and self-review behavior help predict licensure examination performance.
- Ranjeeth, S., Latchoumi, T. P., & Victer Paul, P. The article discusses learning analytics models and their purposes, including monitoring, feedback, forecasting, assessment, personalization, recommendation, mentoring, guidance, and intervention.
- Molerov, D., Federiakin, D., Zlatkin-Troitschanskaia, O., Shenavai, K., Trierweiler, L., & Nagel, M. The study found that students using AI chatbots completed domain-specific critical online reasoning tasks faster, but did not achieve significantly better written performance, and that regular AI users passed more exams but did not necessarily achieve better grades.
- Yaacob, W. F. W., Nasir, S. A. M., Yaacob, W. F. W., & Sobri, N. M. The study compared K-Nearest Neighbor, Naïve Bayes, Decision Tree, and Logistic Regression models for predicting student performance, with Naïve Bayes outperforming the other classifiers based on accuracy, precision, recall, and ROC measures.
