Predictive analytics in schools: a guide for school leaders

What is predictive analytics in education?

Predictive analytics uses patterns in existing data to estimate the likelihood of a future outcome. In a school, that might mean identifying a learner whose attendance, academic trajectory or engagement pattern suggests that support may be needed soon.

The output is a prompt for inquiry, not a diagnosis. A probability, trend or risk band cannot explain family circumstances, health, classroom dynamics or a learner's intent. Those require context from people who know the student.

A useful definition

Predictive analytics helps a school decide where to look sooner. It does not decide what a learner is or what will happen to them.

What data may indicate a need for support?

Useful signals are connected to an outcome, available early enough to act on and capable of changing through support. Schools should prefer a small, defensible set over collecting data simply because it exists.

• Attendance: frequency, recent change, consecutive absence and pattern by day or subject.
• Academics: current performance, rate of change, missing work and differences across subjects.
• Behaviour: documented incidents and changes in participation - interpreted carefully and checked for bias.
• Interventions: what support was offered, when, by whom and what happened next.
• Teacher observations: factual, time-stamped context that structured data may miss.
• Engagement: participation or completion signals that have a clear educational meaning.

Reporting and prediction answer different questions

A dashboard summarises the past and present. A predictive layer looks for patterns associated with a possible future state. Both are useful, but leaders should not confuse a visually sophisticated report with a tested prediction.

How an early-warning prediction works

A responsible workflow starts with a clearly defined outcome and time horizon. Data is checked for quality, relevant indicators are combined, and the system produces a signal that authorised staff can review. The school then adds context, decides whether support is appropriate, records an intervention and follows the learner's progress.

Performance must be monitored over time. A model can drift as curriculum, attendance practices, cohort composition or data-entry behaviour changes. Schools should ask how often results are reviewed, what false positives and false negatives look like, and whether some student groups are affected differently.

1. 1. Current state

   Bring together recent, relevant and reliable signals.

2. 2. Estimate

   Calculate a transparent risk, trend or forecast for a defined horizon.

3. 3. Human review

   Check the evidence and add teacher or pastoral context.

4. 4. Support

   Choose the least intrusive, most appropriate response.

5. 5. Learn

   Monitor progress and evaluate whether the response helped.

What predictive systems should not be used for

Predictive systems should support human decision-making, not label, judge or characterise students. They should not be used to deny opportunity, automate discipline, rank a child's worth or present a forecast as certainty.

Sensitive or proxy variables deserve special scrutiny. Demographic data may be useful for checking whether a system supports groups equitably, but it should not become a shortcut for deciding which individual learner is “risky”. Predictions should be visible only to people who can provide legitimate support.

• No permanent “at-risk” label in the student identity.
• No automated punitive action or high-stakes decision.
• No opaque score that staff cannot interrogate.
• No collection of unrelated personal data “just in case”.
• No intervention without documenting purpose, owner and follow-up.
• No replacement for a teacher, counsellor or safeguarding process.

Questions leaders should ask before adoption

A product demonstration should show the decision workflow, not only the dashboard. Ask the vendor to explain the outcome, time horizon, data used, missing-data treatment, reason codes, access controls and monitoring process in plain language.

• What exactly is being predicted, and how far ahead?
• Which fields influence the result, and why are they educationally relevant?
• Can an authorised user see the contributing signals?
• How are data quality, bias and model drift reviewed?
• Who can access, export or change the record?
• How is an alert resolved, corrected or challenged?
• Can the school measure intervention outcomes separately from prediction accuracy?

Human judgement, privacy and fairness

The safer design is decision support: the system surfaces a pattern, a person reviews it and the school chooses an educational response. UNESCO's guidance on AI in education emphasises human agency, privacy and age-appropriate use. Those principles apply even when the system is predictive rather than generative.

Schools should document a lawful purpose, minimise inputs, limit retention, use role-based access, record important actions and explain the process to staff and families in accessible language. Regular equity checks should compare alert and intervention patterns across appropriate groups without using those group characteristics to stigmatise individuals.