Analysis of alternative graphical representation for the self-organizing mapping of the supersymmetry dataset

Abstract

High energy physics (HEP) simulation and experimentation data are often high dimensional containing high number of features. A beyond standard model (BSM) dataset that is the supersymmetry (SUSY) event simulation dataset was clustered using self-organising map (SOM) algorithm. SOM clustering is one of the better methods to cluster high dimensional data. To verify the existence of the SUSY event in the clustered dataset, it was visualised through several different methods which are the U-matrix, principal component analysis (PCA) and spectral graph theory. U-matrix is the default representation of SOM that visualises the distance between SOM neurons. PCA reduces the dimensionality of the dataset to only 2-D and 3-D considering only the principal components. Spectral graph connects all the neurons together as a network but the implementation was limited by computational resources due to connecting all the neurons of the high dimensional data requires much more intense computational power. While both U-matrix and PCA are successful in visualising cluster(s) in digit datasets, U-matrix was unsuccessful in showing cluster for the SUSY dataset. PCA on the other hand manages to display cluster existence in the SUSY dataset. This may suggest that U-matrix is limited to a certain number of dimensions and PCA might be a better option for cluster existence verification. Further research needs to be done to probe into the potential of dimensionality reduction of clustered HEP data. The visualisation of cluster existence hints to the potential of the algorithm to be used on actual experimentation dataset.