AI Telematics Fuel Efficiency Model
Budget: $250 – $750 USD
I have a sizeable, time-stamped collection of vehicle telematics that combines three rich streams:
• GPS data
• Engine data
• Driving behavior data
My sole objective is to turn these inputs into an accurate, production-ready regression model that can predict and ultimately improve fuel efficiency across our fleet. The data already sits in clean CSV batches, each trip keyed by vehicle ID and timestamp; your job is to move from raw files to actionable predictions.
What I need from you
• Build an end-to-end data pipeline that ingests, cleans, and engineers features from the three datasets above, handling missing values and synchronizing timestamps.
• Select and compare appropriate regression techniques—linear, elastic-net, gradient boosting, or any other approach that demonstrates superior performance—using sound cross-validation.
• Produce clear metrics (MAE, RMSE, R²) and a short report explaining which variables most influence fuel use so my operations team can act on them.
• Deliver well-commented Python code (Jupyter notebook or .py scripts), a requirements.txt, and a brief README so the model can be reproduced internally.
Acceptance criteria
1. Model meets or beats a baseline of ±5 % error on a blind test set I will provide after initial training.
2. All preprocessing and training steps are fully reproducible on a fresh machine with the supplied instructions.
3. Feature importance or SHAP analysis is included to make results interpretable for non-data-scientists.
If you have previous experience with telematics or large-scale sensor data, highlight it in your proposal along with example metrics you achieved. I’m ready to share a data sample as soon as we agree on the approach and timeline.
• GPS data
• Engine data
• Driving behavior data
My sole objective is to turn these inputs into an accurate, production-ready regression model that can predict and ultimately improve fuel efficiency across our fleet. The data already sits in clean CSV batches, each trip keyed by vehicle ID and timestamp; your job is to move from raw files to actionable predictions.
What I need from you
• Build an end-to-end data pipeline that ingests, cleans, and engineers features from the three datasets above, handling missing values and synchronizing timestamps.
• Select and compare appropriate regression techniques—linear, elastic-net, gradient boosting, or any other approach that demonstrates superior performance—using sound cross-validation.
• Produce clear metrics (MAE, RMSE, R²) and a short report explaining which variables most influence fuel use so my operations team can act on them.
• Deliver well-commented Python code (Jupyter notebook or .py scripts), a requirements.txt, and a brief README so the model can be reproduced internally.
Acceptance criteria
1. Model meets or beats a baseline of ±5 % error on a blind test set I will provide after initial training.
2. All preprocessing and training steps are fully reproducible on a fresh machine with the supplied instructions.
3. Feature importance or SHAP analysis is included to make results interpretable for non-data-scientists.
If you have previous experience with telematics or large-scale sensor data, highlight it in your proposal along with example metrics you achieved. I’m ready to share a data sample as soon as we agree on the approach and timeline.