Programme

Validation in multivariate data analysis and machine learning – common pitfalls and anecdotes

Frank Westad

Validation is an important aspect in science. With a new generation of data scientists that has access to all methods for machine learning for exploratory analysis, classification and prediction, including what falls under the term “AI”, this topic is more relevant than ever. One would think that just “pressing the button”; preprocess data, calculate 1000 models and select the best one is a valid procedure. However, there are several pitfalls with this approach. This leads to a large percentage of scientific publications where the findings are wrong, too optimistic or cannot be reproduced. Furthermore, improper use of methods leads to unrealistic estimates of model performance for real-life applications in industry. This presentation will be a mix of anecdotes and best practices related to model validation.

Multivariate Calibration

Tom Fearn

One of the big success stories in chemometrics has been the enabling of quantitative spectroscopy through the use of multivariate calibration. The talk will describe the problem, a regression with large numbers of correlated predictors, and outline some of the solutions. There will be a bit of history, some chemometric methodology, some warnings about what not to do, and some comments on the future.

Random Forest

Stephan Seifert

Random forest (RF) is a machine learning approach that is increasingly used in chemometrics, especially for the analysis of high-dimensional data. RF consists of a large number of individual binary decisions and has many advantages, such as flexibility in terms of input and output variables and the possibility of internal validation. In addition, RF can generate importance measures that can be used to select relevant features.  This presentation explains the basic principles of RF.

"Chemometrics, AI, and Co - Not Just a Topic for Universities: A Report from the Official Food Control Laboratory Landscape

Jo Riehle

Official governmental labs have been considered quite old-fashioned and slow over the last few decades. However, the reality is that these laboratories have also had to meet modern requirements. Data management, short reaction times, and financial limitations are facing young, highly qualified 'digital natives' equipped with up-to-date technical equipment that easily competes with private laboratories. Digitalization and automation are no longer foreign concepts.

With the freedom to search for non-targeted compounds and substances that truly matter, official labs now use modern methodologies such as AI and chemometrics. This presentation discusses the practical opportunities that modern technologies like artificial intelligence offer in the official food control laboratory landscape."