← Back to projects

Dec 2024 - Present

Driver Efficiency Scoring with Double Machine Learning

Built a causal evaluation framework that separated driver behavior from route and weather confounders, producing fair rankings and actionable coaching insights for fleet operations.

PythonPyTorchScikit-LearnSHAPAWS S3Statistical Inference

Drivers Evaluated

35+

Efficiency Gap Quantified

$0.14 kWh/km (12%)

Savings Opportunity

$38K+/fleet annually

Fairness Stabilization Trend

Normalized confounder-adjusted scoring performance over release cycles.

S1S2S3S4S5S6
Drivers evaluated35+
Efficiency gap quantified12%
Fleet savings opportunity$38K+/year

As confounder controls improved, ranking consistency increased and recommendations became easier to operationalize.

Causal DAG

causal_dag.py — DML Driver Efficiency
Confounder
Treatment
Mediator
Outcome
Instrument
· click to expand ·

Problem

Fleet performance comparisons were noisy because route topography, traffic, and weather heavily biased raw efficiency measurements.

Approach

I designed a Double Machine Learning pipeline to isolate behavioral effects from environmental covariates.

  • Gathered and cleaned telemetry from 55K+ trips, then engineered route and driver features for fair comparison.
  • Built treatment and outcome models with cross-fitting to reduce overfitting bias.
  • Added mixed-effects modeling so scores remained stable across driver cohorts and repeated trips.
  • Introduced matched-pair validation and SHAP diagnostics to confirm directionality and model consistency.

Outcome

The final scorecard highlighted a 12% driver-efficiency gap across 35+ drivers with statistically defensible confidence, enabling targeted coaching for throttle smoothness and route-specific behavior.

Why it mattered

This work moved discussions from anecdotal feedback to causal evidence and quantified a $38K+/fleet/year savings opportunity, improving trust from operations stakeholders and accelerating coaching adoption.