Statlog - Identifying students at greater risk of dropout

Societal issues

Identifying students at greater risk of dropout

Background

  • All high schools seek to increase their graduation rates.
  • Limited human resources and budgets require interventions to be efficiently targeted.
  • A school board wanted to assess the potential of machine-learning methods for identifying students who were most likely to drop out.
  • They also wanted to use econometric methods to understand what were the most important predictors of dropout.

Our approach

  • We engineered more than 450 features covering sociodemographic characteristics, earning difficulties, attendance, behaviour, and academic results.
  • Several machine learning models were trained and econometric models estimated.

Our results


86%

accuracy in predicting dropout among students.

These tools allowed the school board to proactively allocate resources to students with the greatest needs and risks.