Fleet Fuel Efficiency Prediction Model (v3)

Performance

• Test R2: 0.838
• 5-fold CV R2: 0.808 +/- 0.025
• MAE: 0.191 km/L
• Overfit gap: 0.015

Data

• 6 months (Jan-Jun 2026)
• 56638 clean trips after dedup + validity + IQR outlier removal

Features (37)

Speed Bands (% of total distance)

• speed_0_16_pct
• speed_16_32_pct
• speed_32_48_pct
• speed_48_64_pct
• speed_64_80_pct
• speed_80_96_pct

Weight

• weight_median_kg

Acceleration Zones (% of distance)

• accel_harsh_brake_pct_dist
• accel_moderate_brake_pct_dist
• accel_light_decel_pct_dist
• accel_cruising_pct_dist
• accel_light_accel_pct_dist
• accel_moderate_accel_pct_dist
• accel_harsh_accel_pct_dist

Torque Zones (% of distance)

• torque_very_low_torque_pct_dist
• torque_low_torque_pct_dist
• torque_mid_torque_pct_dist
• torque_high_torque_pct_dist
• torque_very_high_torque_pct_dist

Idle

• idle_pct_duration
• idle_fuel_pct

Coasting

• coasting_pct_distance

Cruise Control

• cruise_pct_distance

Green Area

• green_area_pct_distance

Time of Day (% of distance)

• dist_pct_00_04
• dist_pct_04_08
• dist_pct_08_12
• dist_pct_12_16
• dist_pct_16_20
• dist_pct_20_24

Braking

• brake_count

Altitude (15-min smoothed)

• pct_dist_flat_v2
• pct_dist_climb_mild_v2
• pct_dist_climb_steep_v2
• pct_dist_descent_mild_v2
• pct_dist_descent_steep_v2
• total_altitude_gain_v2

Driving ratio

• driving_ratio

Example Payload

{
  "speed_0_16_pct": 1.4,
  "speed_16_32_pct": 6.9,
  "speed_32_48_pct": 15.3,
  "speed_48_64_pct": 65.0,
  "speed_64_80_pct": 10.9,
  "speed_80_96_pct": 0.5,
  "weight_median_kg": 28500,
  "accel_harsh_brake_pct_dist": 1.2,
  "accel_moderate_brake_pct_dist": 4.1,
  "accel_light_decel_pct_dist": 28.3,
  "accel_cruising_pct_dist": 30.5,
  "accel_light_accel_pct_dist": 31.2,
  "accel_moderate_accel_pct_dist": 4.0,
  "accel_harsh_accel_pct_dist": 0.4,
  "torque_very_low_torque_pct_dist": 46.0,
  "torque_low_torque_pct_dist": 38.5,
  "torque_mid_torque_pct_dist": 12.0,
  "torque_high_torque_pct_dist": 2.5,
  "torque_very_high_torque_pct_dist": 0.3,
  "idle_pct_duration": 4.8,
  "idle_fuel_pct": 0.95,
  "coasting_pct_distance": 16.0,
  "cruise_pct_distance": 1.2,
  "green_area_pct_distance": 79.5,
  "dist_pct_00_04": 15.0,
  "dist_pct_04_08": 25.0,
  "dist_pct_08_12": 30.0,
  "dist_pct_12_16": 20.0,
  "dist_pct_16_20": 8.0,
  "dist_pct_20_24": 2.0,
  "brake_count": 650,
  "pct_dist_flat_v2": 91.6,
  "pct_dist_climb_mild_v2": 3.7,
  "pct_dist_climb_steep_v2": 0.3,
  "pct_dist_descent_mild_v2": 3.1,
  "pct_dist_descent_steep_v2": 1.3,
  "total_altitude_gain_v2": 1200,
  "driving_ratio": 0.89
}

Usage

from xgboost import XGBRegressor
import json

model = XGBRegressor()
model.load_model('fleet_fe_xgboost.json')

with open('feature_cols.json') as f:
    feature_cols = json.load(f)

prediction = model.predict(trip_features[feature_cols])

Changelog

• v3: Speed bands as %, smoothed altitude features (15-min rolling median), CV R2=0.808
• v2: Raw speed bands, 6-month data, CV R2=0.796
• v1: Initial model, May only