AI volume prediction error across Durham census tracts. Darker red indicates higher prediction errors, concentrated in low-income areas.
Prediction accuracy across income quintiles (Q1=poorest, Q5=richest). Shows mean absolute error in predicted vs actual pedestrian/cyclist counts.
Prediction errors grouped by census tract racial composition. Areas with higher minority percentages show systematically worse accuracy.
Predicted volumes vs actual counts. Perfect predictions follow the diagonal. Systematic deviations reveal where bias occurs.
Each dot is one counter location. Dots left of zero indicate underprediction by the AI model.
Actual vs predicted crash counts by census tract. Toggle between views to compare.
Binary classification (above/below within-quintile median) evaluated per income group. Lower scores in poorer quintiles indicate the model struggles to rank tracts within those areas.
Actual vs predicted crashes over time. Shows persistent over/underprediction patterns by income level.
Relative prediction error by income level. Higher error in poorer quintiles indicates systematic bias in model accuracy.
Safety project locations from AI vs need-based allocation. Shading shows danger scores; markers show projects. Toggle to compare.
AI-driven vs need-based safety budget allocation per income quintile.
Four normalized equity metrics (0-100) comparing AI-driven and need-based allocation strategies.
Suppressed, potential, and actual cycling/walking demand across Durham. High suppression (red) indicates latent demand AI tools miss.
Demand suppression stages from potential to actual usage. Width represents trip volume at each stage. Q1 areas show severe drop-off.
Detection accuracy for suppressed demand. Naive AI fails; sophisticated AI achieves partial detection. Neither matches human expert baseline.