useR! 2024

The widespread emergence of object data, such as images, text, graphs, matrices, etc., emphasizes the need for new data analysis methods tailored to these unusual data formats. This is a difficult task as even a simple operation such as computing the mean (the most central point) is no longer straightforward in object data which have a very complex structure. Moreover, in order to ensure that our analysis stay reliable, it is imperative to develop robust methods, which are tolerant against outliers regardless of their exact type. In this talk we show how different robust depth functions can be used in R to find the most central objects in any object data set. As a byproduct, we also identify any possible outliers in the data. We offer two alternatives for the estimation: a fast in-sample optimization and a slower and more accurate out-of-sample version based on the evolutionary algorithms in the R package "GA". The method is particularly useful in the descriptive analysis of object data, which allows to detect outliers and find what the most typical objects look like.