Machine learning in logistics: Make every minute count

Press Review

back to overview

This article was published in the September issue of Global Cement Magazine in 2020.

Logistics in the cement, ready-mix concrete and aggregates industry is facing a major overhaul as traditional tools and processes are enhanced or even replaced with machine learning (ML). When it comes to automated decision-making in transport planning, ML algorithms can move predictions one step closer to reality.

We have all heard a lot of hype and excitement around machine learning (ML). However, there’s a lot of confusion about what the term ML really means and what it can do today. So, before we take a look at how it can improve the logistics performance in our industry, let’s break down fi ve of the common myths around ML.

by Dr. Paul Flachskampf and Dirk Schlemper, INFORM GmbH, Germany

Myth 1: ML can do anything with data. For many, ML simply means shovelling piles of data into one side of a computer and collecting ‘answers’ on the other side. What if the answers are wrong? Well, just stir the pile until the answers start looking right. Right?

No, it isn’t that easy. ML is a sophisticated set of technologies that allows you to use the data generated by your business. Before you can expect answers, however, you have to understand the problem that you are trying to solve.

Myth 2: ML and data mining is ‘easy.’ Early experiences in ML oft en resulted in disappointing results because analysts grabbed data sources without vetting them fi rst. Before taking any action, data has to be clean and accurate. To benefi t from big ‘data lakes,’ it is important to truly understand their sources. Where does it come from? Who has manipulated it? Are they reliable? At the same time, make sure that you have enough data to discover the patterns and anomalies within that data. ML cannot turn missing, incomplete, or wrong data into valuable insights.

Myth 3: ML is about predicting the future. This is true... and false. Keep in mind that ML will always be trained on historical data, so it will struggle to predict the future in situations where the future is expected to diverge from the past, e.g.: predicting the next pandemic crisis. Instead, you should use ML for its full range of use cases: to generate business insights and to add new application features, in addition to predicting outcomes and forecasting.

Continue reading - request the full article using the form below.

Data Protection*
Back to top