Predicting Demand for Commodities using Machine Learning

Outline:

Time-series prediction is often considered a niche domain in machine learning; it can be less intuitive to understand and more complex to model than simple classification and regression models.

But, all natural and human processes can in some way be represented as a time-series, everything from the movement of the polar ice shelf, to how gazelle move through North America, even to how you used Tinder after your last breakup to find a bounce-back relationship.

Time-series analysis is not only important to understand, it is now becoming the top application of machine learning in many industries. For example, in manufacturing, predictive maintenance driven by time-series prediction of machine outages has added millions to the bottom line of major manufacturers.

In this presentation, we'll:

• Cover a real-life application of time-series prediction - forecasting demand for a specific commodity

• Go through the business applications, cost savings, and technical methods behind the work

Speaker's Bio:

Lou Zhang is a Data Analyst on the Advanced Analytics team at IHS Markit. Lou develops predictive models for numerous industrial applications - forecasting car sales, commodity prices, and supply and demand curves for key industrial materials. Lou has previous experience in economic consulting and product management -- experience which is helpful in the world of machine learning when he needs to transform legacy econometric models into modern machine learning models.