Understanding the behaviour of metaheuristics is an important step towards their explainability. This, in turn, is vital for multiple reasons. Not only can researchers use those insights to improve existing metaheuristics, e.g. by inventing new operators, but it also allows those researchers to even be aware of what parts merit investigation and can therefore aid in study design. Additionally and perhaps most crucially, being able to provide practitioners clear behavioural patterns for algorithm configurations can vastly increase the number of people that can successfully use metaheuristics for their optimization needs without having to deeply understand how to select and combine operators or how to set certain hyperparameters. In this paper, we propose the use of unsupervised clustering to make behaviour analysis easier and more feasible for large-scale experiments. Building on previous work, we reintroduce a number of metrics and propose the use of meaningful features for analysis rather than entire behaviour chains. We demonstrate on the example of convergence curves how even simple clustering techniques can be used to sort a large number of different behaviours into groups, which decreases the need for manually analysing individual outcomes by orders of magnitude in our experiment with more than ten thousand configurations.