Hi Nils,
I understand that you want to perform cluster analysis on multi-dimensional matrices. You could do so by extending the ‘k-means’ clustering algorithm to work on this multi-dimensional matrix. The cluster shapes would be multi-dimensional with a multi-dimensional point as its centre and standard Euclidean distance as its radius. You could also define the clusters with other metrics and distances such as ‘mahalanobis distance’ or ‘cosine similarity’ as per requirements. You can then extend the standard ‘k-means’ algorithm to work with these modified clusters and data points.
For example, cluster analysis of 3-D data can be done using the above technique. You could define spherical clusters with radius as a parameter or cuboidal clusters with parameters length, breadth, and height and then extend the ‘k-means’ algorithm to work on this 3-D data.
To read more about k-means clustering, please refer the following documentation: https://en.wikipedia.org/wiki/K-means_clustering
To read more about clustering using MATLAB, please refer the following documentation: https://www.mathworks.com/help/images/color-based-segmentation-using-k-means-clustering.html
I hope this helps.
