I will build a time series forecasting model for energy and sensor data

Most forecasting models look accurate in backtest and fail in production. The reason is almost always the same: random cross-validation on time-ordered data, and a single model trained across regimes that behave differently.

I build forecasting models on time series and sensor data, validated the way they will actually be used.

WHAT YOU GET

- A model trained and validated on strictly chronological splits, never shuffled

- Honest accuracy metrics (MAE, RMSE, R²), reported per regime rather than as a single flattering average

- A clear statement of where the model is reliable and where it is not

MY REFERENCE PROJECT

A 24-hour wind power forecasting system running in production, validated on two independent sites with different wind climates and a regime-calibrated architecture. See my portfolio for the measured results.

TYPICAL USE CASES

Energy demand and generation, industrial sensor readings, equipment and load prediction, any signal where time ordering matters.

WORKS BEST WITH

Historical data with timestamps, ideally 6+ months. Message me with your dataset size and target before ordering and I will tell you honestly whether it is enough.