Predicting Student College Outcomes — Empower Schools

Built an SVM classifier predicting student college outcomes with 80% accuracy across 150,000+ students in the Lawrence and Springfield, MA school districts. Presented findings directly to the Empower Schools founder and executive team.

Summer data-science research project investigating accountability and performance data across the Lawrence and Springfield, MA school districts.

Scope

• 150,000+ students in the dataset.
• Investigated correlations between student outcomes, GPA, standardized-testing scores, and demographic variables.
• Built predictive models of student success using Python ML.

Result

A support-vector-machine classifier that predicted student college outcomes with 80% accuracy, presented directly to the Empower Schools CEO/founder and the executive staff. The classifier itself isn’t the novel piece — it’s a reasonable applied-ML result for this dataset shape — but the surrounding work, sourcing the data, defining the prediction target carefully enough to be useful, and presenting the result to non-technical decision-makers, was where most of the value lived.

This was also where I first noticed how much weight predicted-outcome models can carry in real institutional decisions, and how careful you have to be about who’s actually being predicted.