<p>Space debris around the Earth are becoming an increasing threat for space missions. Their number is growing due to the frequent launches of satellites from space agencies and private enterprises. This study examines simulated break-up events alongside actual samples of catastrophic events; we analyse the generated fragments with the objective of assigning them to clusters and classifying the debris based on their dynamical properties. We propose to accomplish these goals by performing the analysis using the so-called proper elements, which are quantities obtained by implementing perturbation theory to average the equations of motion over the angle variables. Subsequent to this filtering procedure, the proper elements enjoy the remarkable property to remain nearly constant over time. We find that proper elements are highly suitable for the analysis through machine learning methods with the purposes of clustering and classifying the fragments.</p>