AI Model Drift in Education

This dataset provides detailed tracking of AI/ML model drift events and associated performance changes in educational software, including context such as classroom type, institution, and remediation actions. It enables edtech teams to monitor, audit, and improve predictive models for student learning outcomes in remote and hybrid classrooms, supporting transparency and continuous model improvement.

  • last updated 11 Jan 2026
  • by GoMask
The brief that made it

Monitoring and auditing AI/ML model drift in educational platforms

Sample rows

preview · 8 of 50 rows · all 19 columns
drift_event_idstringdrift_typestringdrift_severityfloatmodel_namestringmodel_idstringmodel_versionstringinstitution_idstringinstitution_namestringclassroom_typestringdrift_detected_atdatetimeperformance_metricstringmetric_value_beforefloatmetric_value_afterfloatdetected_bystringremediation_actionstringremediation_timestampdatetimelearning_outcome_impactstringnotesstringdrift_descriptionstring
DRFT-001concept0.73EduPredictorM-1001v2.3.1INST-101Greenwood Highremote2023-09-14T09:23:00Zaccuracy0.920.77auto-monitoringretrain2023-09-15T11:10:00ZdegradedRetraining scheduled after detection.Decrease in accuracy after curriculum update.
DRFT-002performance0.62OutcomeInsightM-1002v1.9.0INST-102Cedar Valley Collegehybrid2023-10-01T14:45:00Zf1_score0.880.81user-reportalert2023-10-01T15:15:00ZdegradedAlert sent to model admin.Model performance dropped after exam week.
DRFT-003data0.29StudentSuccessAIM-1003v3.0.2INST-103Eastside STEM Academyin-person2024-01-08T08:20:00Zprecision0.840.8system-checkignoreblankunchangedMinor drift, action not needed.New attendance data introduced anomalies.
DRFT-004performance0.54EduPredictorM-1001v2.3.1INST-104Lakeshore Institutehybrid2023-11-04T13:10:00Zrmse1.021.58auto-monitoringretrain2023-11-05T12:30:00ZdegradedModel retrained with recent data.Unexpected increase in RMSE post holiday break.
DRFT-005data0.41LearnTrendM-1004v4.0.0INST-105Hilltop Academyremote2023-08-19T10:50:00Zrecall0.760.72teacher-reportalert2023-08-19T11:00:00ZdegradedAlert sent to technical support.Recall dropped due to change in grading system.
DRFT-006concept0.58PathwaysAIM-1005v2.5.1INST-106Pacific Learning Hubin-person2023-12-04T16:30:00Zauc0.810.77system-checkrollback2023-12-04T17:45:00ZunchangedRolled back to previous model version.AUC decreased after new subject introduction.
DRFT-007performance0.37StudentSuccessAIM-1003v3.0.2INST-101Greenwood Highremote2023-09-22T10:15:00Zaccuracy0.90.83user-reportalert2023-09-22T10:30:00ZdegradedAdmin alerted for further investigation.Accuracy decreased after student transfer.
DRFT-008concept0.65ProgressTrackM-1006v1.2.3INST-107Elmwood Academyhybrid2023-08-30T15:25:00Zprecision0.880.74auto-monitoringretrain2023-08-31T09:00:00ZdegradedModel retrained on new format.Precision dropped due to modified test format.

What the 50 rows show

from the 50-row sample

Data (drift type) stands out: mean drift_severity is 0.28, against 0.59 for the rest.

  • 0.52median drift_severity
  • 3classroom types
  • 3learning outcome impacts
  • 4detected_by values
  • 4remediation actions
  • 6performance metrics
Mean drift_severity by drift_type50 rows
00.40.80.28data15 rows0.67concept15 rows0.53performance20 rows
drift_severity50 rows, in bands of 0.1
05101106799530.10.50.9drift_severity →

Median 0.52, from 0.18 to 0.84.

model_name50 rows · 6 values
  1. PathwaysAI10
  2. ProgressTrack10
  3. OutcomeInsight8
  4. LearnTrend8
  5. EduPredictor7
  6. StudentSuccessAI7
19 columns by typefrom the column list below
  • string 14
  • float 3
  • datetime 2

Columns

19 columns in three groups
blueprint · 19 columns
columntypedescriptionexample
Text 14 columns
drift_event_idstringUnique identifier for each model drift event.uniqueDRFT-001
model_idstringUnique identifier for the AI/ML model experiencing drift.M-1001
model_namestringName of the AI/ML model.6 namesEduPredictor
model_versionstringVersion of the AI/ML model.v2.3.1
institution_idstringUnique identifier for the educational institution using the model.INST-101
institution_namestringName of the educational institution.Greenwood High
classroom_typestringType of classroom environment (e.g., remote, hybrid, in-person).remote · hybrid · in-personremote
drift_typestringType of drift detected (e.g., data drift, concept drift, performance drift).data · concept · performanceconcept
performance_metricstringName of the key performance metric affected (e.g., accuracy, F1 score, RMSE).6 valuesaccuracy
drift_descriptionstringDetailed description of the drift event and its context.optionalDecrease in accuracy afte…
detected_bystringName or identifier of the system or person who detected the drift.4 values · optionalauto-monitoring
remediation_actionstringAction taken to address the drift event (e.g., retrain, rollback, alert, ignore).retrain · rollback · alert · ignore · other · optionalretrain
learning_outcome_impactstringAssessment of the drift's impact on predicted student learning outcomes (e.g., improved, degraded, unchanged, unknown).improved · degraded · unchanged · unknown · optionaldegraded
notesstringAdditional notes or comments related to the drift event.optionalNo action required.
Numbers 3 columns
drift_severityfloatSeverity score of the drift event (e.g., 0.0-1.0).0 to 10.73
metric_value_beforefloatValue of the performance metric before drift was detected.0 or more0.92
metric_value_afterfloatValue of the performance metric after drift was detected.0 or more0.77
Dates and times 2 columns
drift_detected_atdatetimeTimestamp when model drift was detected.2023-09-14T09:23:00Z
remediation_timestampdatetimeTimestamp when remediation action was applied.optional2023-09-15T11:10:00Z

Use it for

  • median drift s…0.5250 rowsmean drift severity b…0.28data0.67concept0.53perfor…

    An education dashboard

    Drift_severity by drift_type and a breakdown of model_name. Excel, Power BI or Tableau.

  • Why do the 15 data rows have a mean drift_severity of 0.28?

    A root-cause class exercise

    Hand out the rows and one question. The answer is in the data, not in the brief.

  • A software demo

    Believable drift events with model_id, model_name and model_version to fill a screen in front of a buyer.

Not quite right?

Make it yours.

Same 19 columns, your size and your rules. See 20 rows before you pay.

Preview 20 rows free

1,000 rows: $1.99 · 10,000: $12.99One-time. No subscription. All prices

This dataset50 rows19 columns
Yours1,000 rows19 columns

blueprint · ai-model-drift-in-education

Behind this dataset

Same schema. As many rows as you need.

These 50 rows came out of a blueprint — 19 columns with generation rules behind each one. Open it in Data Factory to retune a column, add your own, wire in foreign keys, and run it at the size you actually need.

Rules it was built with
  • Each row logs a model drift event with timestamp, model version, and detected performance drop.
  • Include root cause analysis and corrective actions taken for each drift event.
  • Log associated student group, subject area, and deployment environment (remote/hybrid/classroom).
  • Performance metrics before and after drift event must be recorded.
  • Flag critical drifts impacting assessment accuracy.
Rows
Open the blueprint in Data Factory

1 credit per row. New accounts start with 25 free credits.

Exports
CSV, JSON, JSONL, Parquet, SQL, Excel, TSV, XML
Licence
yours to use, including commercially
API slug
ai-model-drift-in-education

More in Education

Browse all
Also in EducationVirtual Classroom Engagement MetricsAdaptive Learning Pathways DatasetOnline Learning Quiz Attempt LogsAll 1,954 datasets

What should your data show?

Preview 20 rows free
No signup. No card.