Event Details

Industrial Problem Solving Day: Tesco - Machine Learning Predictions and Optimisation: Fuel Pricing

Machine Learning Predictions and Optimisation: Fuel Pricing

The data science team at Tesco create prescriptive decision support and automation tools for use across the whole business from shop floor to online deliveries. Across multiple projects a common theme is emerging; machine learning (ML) techniques are used to make predictions, the predictions are then used within an optimisation model to arrive at a recommended decision for the business.

There has been significant work done on using optimisation techniques within machine learning algorithms [1]. However in the words of Deng et al. “ML provides support for predictive analytics, whereas, optimization forms the basis for prescriptive analytics, and the methodologies for these are built (some-what) independently of each other” [2]. One of the challenges for practical applications combining both ML and optimisation is the large number of predictions required by the optimiser, each of which will have a different confidence interval.

In this problem solving day we want to explore approaches that could be taken to fuel pricing. Tesco is the UKs largest seller of fuel through petrol stations. It is a competitive market with little differentiation between brands and so once we change a price competitors can and will respond. The plan is to use machine learning to predict sales given a price and optimisation techniques to maximise profit.

[1] Tirtahjyoti Sarker, “What lies beneath? Optimization at the heart of Machine Learning.” Towards Data Science, Nov 2018, https://towardsdatascience.com/a-quick-overview-of-optimization-models-for-machine-learning-and-statistics-38e3a7d13138

[2] Deng et al. Coalescing Data and Decision Sciences for Analytics, 2018 INFORMS Annual Meeting - Tutorials in Operations Research