Global dairy industry leader applies machine learning
“We relatively quickly could drop out a number of the ingredients we had been testing… This wasn’t obvious if you just looked at them one-by-one, because you always have some cross-interactions. This was a big learning and helped speed up the development.”
Matthias Eisner, Innovation Manager, Yili
- Machine learning was able to provide valuable insights into the behaviour of cream formulations over time.
- Yili was able to remove some ingredients from consideration, speeding up development.
- The project identified useful cross-correlations and was able to predict likely behaviour in later months from earlier test results.
Summary
Global dairy products leader, Yili, applied Alchemite™ machine learning to study UHT whipping cream formulations – a commercially-significant product-line for food services such as large-scale bakery operations. The product must show reliable properties over a nine month shelf life, achieved by using ingredients including stabilisers and emulsifiers, and by controlling processing. The project team built a reliable machine learning model based on 2-3 years of time series formulations data, and showed how this model could be evolved and applied as new data was continuously added. Missing data was imputed and a hierarchical modelling approach was employed to allow earlier months’ tests to be used as inputs to improve predictions of later shelf life. A key learning was understanding which ingredients impacted target properties, enabling some additives to be dropped, speeding up the development process.
