Abstract
Compact steep-spectrum sources (CSS), high frequency peakers (HFP), and gigahertz-peaked spectrum sources (GPS) are compact radio sources with an intense emission (O'Dea 1998, and references therein). Morphological studies, dense gas analyses, and surveys suggesting the absence of a halo diffusion emission, suggest the idea that peaked spectrum sources (PSS) are young AGN (see, Fanti et al. 1995; Readhead et al. 1996; Stanghellini et al. 1997; Bicknell et al. 1997). Previously, Torniainen et al. (2008) carried out a study of GPS sources, finding that those sources do not follow a distinct morphological classification. In addition, they found that many blazars in a flaring state are misclassified as GPS sources (Torniainen et al. 2005). These findings compromised the simple vision of the g alaxy-quasar dualism, and the amount of genuine GPS sources. For this reason, we endeavour a new classification of 363 sources with the aim to find new insights about their spectral properties using neural networks. Through clustering methods, we have grouped galaxies that present a set of similar physical properties. In total, 18 physical variables were used for this purpose. In particular, we used Multi-dimensional Scaling (MDS) and t-distributed Stochastic Neighbour Embedding (t-SNE) analyses. Those analyses proved to be robust for analysing data of the order of hundreds of data. From our analyses, we were unable to find a clear classification for PSS. Moreover, new galaxies presented a flat spectra emission, or high variability in the radio emission, compromising their classification as genuine PSS.
