CUNY Bus Route Analysis - ACE Implementation Study

Project Overview

Research Objective

Quantify the impact of Automated Camera Enforcement (ACE) implementation on CUNY student commute times and identify optimal routes for deployment based on route characteristics, stop patterns, and urban context.

Data Sources

MTA GTFS data with detailed route geometry and stop patterns, NYC traffic violation database, comprehensive CUNY student survey with 185 validated responses across all boroughs.

Machine Learning

33 Routes

Ridge Regression model trained on historical ACE deployments using static route characteristics

Methodology & Approach

Data Collection and Processing

MTA GTFS Data: Analyzed detailed route geometry, stop patterns, and service characteristics for all NYC bus routes. Calculated stops per mile, route directness, turning frequency, and signal priority potential.

ACE Implementation Records: Compiled database of 33 routes with documented ACE deployment from NYC DOT, providing training data for predictive modeling.

Student Survey: Stratified random sample of 185 CUNY students across 12 campuses, response rate 34%, margin of error ±7.2% at 95% confidence level. Survey instrument validated through pilot testing.

Machine Learning Pipeline

Feature Engineering: Created route-level features including stops_per_mile, directness_ratio, turns_per_mile, avg_stop_spacing, close_stops_ratio, signal_priority_potential, and urban_density indicators.

Data Preprocessing: StandardScaler normalization applied to all numeric features. Feature selection using SelectKBest to identify the 8 most informative characteristics.

Model Training: Trained on 33 historical ACE routes using strong regularization appropriate for small sample size. Applied 5-fold cross-validation to assess model stability.

Model Selection: Ridge Regression with strong regularization selected based on cross-validation performance and interpretability of route characteristics.

Model Interpretation: Our model learns from patterns in 33 historical ACE deployments. The predictions indicate which routes share characteristics with previously selected routes. These are directional indicators for further investigation, not precise forecasts of implementation decisions or outcomes.

Student Impact Assessment

Student Survey Results - How often are you late to classes because of your commute

Survey Results Analysis (n=185)

65% of respondents report being late to class sometimes, often, or always due to commute-related delays
32.4% indicate they are occasionally late due to their commute
31.4% report being late sometimes
18.4% are often late to class
15.1% are always late due to transportation issues
Only 3.2% report never experiencing lateness from their commute

65%

of CUNY students experience class lateness due to inadequate bus service

Survey data reveals that transportation delays represent a significant barrier to academic success for CUNY students. Routes M103 and B103 emerged as the most frequently utilized, each mentioned 10 times by respondents. Routes M15, M20, and M9 constitute secondary transportation corridors with substantial student ridership. This utilization data directly informed our prioritization methodology for ACE deployment recommendations.

Most Common Buses Taken to CUNY Campuses

Interactive Route Checker

Check ACE Priority for Your Route

Search any of 220 NYC bus routes to see predicted improvement potential based on machine learning analysis of route characteristics. Predictions are based on patterns from 33 historical ACE deployments and should guide further investigation.

All Routes (220)

CUNY Routes (21)

Top CUNY Routes by Predicted Improvement:
M103 (Hunter) +14.9% B49 (Kingsborough) +14.6% B1 (Kingsborough) +14.7% M22 (BMCC) +14.0% M20 (BMCC) +13.7%

Current Speed

Predicted Improvement

Predicted Speed

Campus

What this means: This prediction is based on route characteristics that historically correlated with ACE deployment. Actual implementation decisions depend on enforcement capacity, street infrastructure, political feasibility, and local conditions not captured in this model.

Compare Routes Side-by-Side

Route Comparison Tool

Select 2-3 routes to compare their current speeds, predicted improvements, and route characteristics. This helps identify which routes would benefit most from ACE implementation.

Route 1 (Required)

Route 2 (Required)

Route 3 (Optional)

ACE Implementation Impact Assessment

Speed Change Distribution ACE vs Non-ACE

Model Performance Highlights

Training Dataset: 33 routes with documented ACE implementations analyzed for route characteristics
Feature Selection: 8 most informative features identified from stop patterns, geometry, and urban context
Cross-Validation: 5-fold CV with strong regularization appropriate for small sample size
Model Interpretation: Stops per mile, signal priority potential, and route directness emerged as key characteristics
Predictions Generated: 220 routes analyzed for ACE deployment priority based on learned patterns

+0.2 mph

Net advantage of CUNY ACE buses vs non-ACE CUNY buses

Analysis shows that ACE technology improved average bus speeds across all routes, with CUNY routes increasing from 7.6 mph to 8.3 mph and non-CUNY routes rising from 8.3 mph to 8.8 mph. The consistent gains demonstrate that ACE delivers meaningful performance improvements throughout the network, with CUNY routes benefiting significantly despite serving more congested corridors with frequent stops.

Priority Routes for ACE Implementation

Machine Learning Prediction Methodology

To identify which routes would most benefit from Automated Camera Enforcement (ACE), we employed machine learning techniques using route characteristics and geometry.

1. Route Feature Engineering

Calculated stops per mile, average stop spacing, and close stops ratio from GTFS data
Measured route directness, turns per mile, and signal priority potential from geometry
Encoded urban density, crosstown characteristics, and route type
Selected 8 most informative features using statistical analysis

2. Model Training & Evaluation

Trained Ridge Regression model on 33 historical ACE deployments:

Ridge Regression: Selected as best model with strong regularization
Training Sample: 33 routes with documented ACE implementation
Validation: 5-fold cross-validation to ensure model stability
Features: Focus on stop patterns, geometry, and urban context

3. Generating Predictions

The model identifies routes sharing characteristics with historically selected ACE routes. Predictions indicate deployment priority based on learned patterns from 33 training examples.