The early warning signals that actually predict a dropout
A student who drops out in the sixth semester did not decide in the sixth semester. The decision usually accumulates over a year or more, through a sequence that is visible in the college's own records long before anyone acts on it — and is almost never assembled into a picture, because each signal sits with a different person.
This is about which signals matter, how early they appear, and the part most colleges get wrong: what to do once you have identified someone.
The signals arrive in a predictable order
Attendance decline usually comes first, and it is gradual rather than sudden — a student moves from ninety per cent to seventy over several weeks. Because no single week looks alarming, monthly reporting tends to miss it entirely; only the trajectory is informative.
Assessment performance follows, typically a term behind. The important pattern is not a low mark but a falling one — a student dropping from consistent sixties to consistent forties is a much stronger signal than a student who has always scored forty.
Disengagement from optional activity comes somewhere alongside. Stopping attending things they are not required to attend is often the earliest visible sign, and the one least likely to be recorded anywhere.
Withdrawal from peers is usually the last visible signal before departure, and by then intervention is much harder.
Trajectories, not thresholds
Most colleges monitor thresholds — flag anyone below seventy-five per cent attendance. Thresholds are administratively convenient and diagnostically poor. They flag students who have always been marginal and miss the student who fell from ninety-two to seventy-eight in six weeks, who is the one actually in trouble.
The signal is the slope. This is difficult to see by hand across hundreds of students, which is precisely why it is worth computing rather than eyeballing.
Detection is the easy half
Most systems that identify at-risk students fail at the next step, because a list of names is not an intervention. Somebody has to have a conversation, and it matters enormously how that conversation starts.
This is why a risk score should never be shown raw to the student it describes. Being told by software that you are high-risk is not motivating; for a student who is already disengaging it is confirmation. The output should reach a HOD or principal who reviews it, decides whether it reflects reality, and approaches the student as a person rather than as a flag.
It also matters that detection connects to something. If the only available response is a warning about attendance shortage, you will identify struggling students accurately and then apply pressure to them, which mostly accelerates the outcome you were trying to prevent.
What to put in place
Capture attendance at the period level so trajectory is computable. Enter internal assessment when it happens so the series is dated. Record interventions, so you learn which conversations worked. And decide, deliberately, who sees a risk signal and who talks to the student — before you turn any of it on.
The takeaway
Predicting a dropout is not the hard part; colleges already hold the data. The hard part is arranging for a human to have a good conversation early, which is a policy decision about who sees what — not a modelling problem.
Questions we get asked
Should students see their own risk assessment?
Not as a raw score. Campus360 routes risk snapshots to a HOD or principal for review and publication precisely because an unmediated label tends to accelerate disengagement rather than reverse it.
How much data does a model like this need?
A meaningful trajectory needs at least a semester of period-level attendance and dated internal assessment. Less than that and you are reading noise.
Is this used for academic decisions?
It should not be, and in Campus360 it is not. It is an early-warning input for human intervention and does not gate any academic outcome.
See Campus360 on your own campus data
A 30-minute walkthrough with your departments, your roles and your questions. No slide deck.