Wholesale fuel demand forecast with machine learning

Wholesale fuel demand forecast with machine learning

Business Challenge

Fuel has to be delivered from refineries to regional depots so that it corresponds most accurately to fuel demand and consumption in areas served by individual depots. This has to be accurate as losses from over- or undersupply are vast. In the past, our client – a Hungarian oil and gas company – was doing this in Microsoft Excel by human input which sometimes resulted in inaccurate forecasts and revenue losses.


Starschema experts created a machine learning-based solution that forecasts daily, weekly, monthly and quarterly fuel demand. The application uses a combination of various time series predictors including auto-regressors, an ensemble of competing models and takes a range of seasonal and trend-like factors (e.g. the general state of the economy, fuel price changes etc.) into account to provide more accurate forecasts.


Model accuracy vastly exceeded the current manual demand planning process based on subject matter experts manually creating forecasts. This allowed our clients to save millions in reallocation and storage costs from day zero.

By | 2019-01-26T14:11:43+00:00 November 24th, 2018|case study, data science, featured|0 Comments

About the Author:

Leave A Comment