Nanoindentation has evolved into a method for performing rapid mechanical property characterisation, allowing for easy collection of vast amounts of data. To get insights, Elastic Modulus and Hardness maps are usually plotted to condense the data into a human-readable way. If the constituents making up the material are different enough in terms of Elastic Modulus and Hardness, their properties can easily be extracted from such maps. If not, the data need to be deconvoluted. For this reason, unsupervised machine learning methods such as the K-means clustering are often used and have been shown to be performant. This even leads to the implementation of such algorithms as applications in standard nanoindenter software, allowing every experimentalist to access machine learning methods easily. The question that this talk tries to answer is why, despite enormous efforts to create meaningful mechanical features utilised in solving the reverse-nanoindentation problem in the past, these features have not yet been used in clustering and classification of mechanical property data. The advantages of such features will be shown in the case of a curated High-Speed Steel Dataset consisting of 3300 individually human-labelled indents. It will be demonstrated using unsupervised and supervised (explainable) machine learning that utilising features from every part of the indentation curve can deconvolute maps which cannot effectively be clustered using the Elastic Modulus and Hardness while additionally gaining insights by explaining the model's prediction using methods based on cooperative game theory.