Spotting Students at Risk: Simple Rules, Early Signals

There is a temptation, whenever "at-risk students" comes up, to reach for something sophisticated — a model, a predictive score, a number between zero and one that claims to know who is in trouble. It sounds impressive. It is also, for most schools, exactly the wrong tool, because the moment a teacher asks "why is this student flagged?" the clever answer is a shrug. A score nobody can explain is a score nobody should act on.

Classable takes the opposite approach on purpose. The at-risk signal is rule-based. It is a short list of plain conditions, written in language anyone can read, and a student is flagged when they meet one. There is no machine learning here, no prediction, no hidden weighting. Just rules.

The rules, stated plainly

A student is surfaced as at risk when a simple, visible condition is met:

• Attendance below 80 percent. The threshold that is widely used to mark chronic absence. Below it, a student is missing enough school for the gaps to start compounding.
• Three or more consecutive absences. A run of days, not scattered ones. A consecutive streak often signals something acute — illness, a crisis, the start of a student withdrawing — and warrants a same-week look.
• A below-pass grade. A result under the passing mark is a concrete academic signal, independent of attendance, that a student needs attention.

That is the whole logic. Any one condition raises the flag. A school reading this list knows exactly what the system is watching for, and exactly why a given student appeared.

Why transparency beats cleverness here

The case for rules over a model is not nostalgia. It is that an early-warning signal is only useful if someone acts on it, and people act on what they understand.

• It is explainable. A form tutor can see at a glance that a student is flagged because attendance dropped under 80 percent — and knows immediately what the conversation is about.
• It is trustworthy. A rule cannot surprise you with a verdict it cannot justify. There is no risk of a hidden bias baked into training data, because there is no training data.
• It is defensible. When a school acts on a flag — a call home, a meeting — it can state the reason in a sentence. "She has missed three days running" is a fact. A risk score is an assertion.
• It is tunable by understanding. Everyone involved can reason about what the thresholds mean, rather than treating the output as an oracle.

A black-box score might, in theory, be marginally more accurate. In practice, a signal that staff trust and act on beats a cleverer one they quietly ignore because they cannot explain it.

From flag to conversation

A flag is not a diagnosis, and the rules do not pretend to be. What they do is surface the handful of students who have crossed a line, on the dashboard, so that a human notices early and asks a question while it still helps. The rule decides who appears; the person decides what to do. That division is deliberate — detection is where consistency matters, and response is where judgement belongs.

Classable surfaces at-risk students on the dashboard using exactly these rules — attendance under 80 percent, three or more consecutive absences, a below-pass grade — so a school gets an early, honest signal it can read and trust, not a prediction it has to take on faith.