Active galactic nuclei synapses: X-ray versus optical classifications using artificial neural networks
(Abridged) Many classes of active galactic nuclei (AGN) have been defined entirely throughout optical wavelengths while the X-ray spectra have been very useful to investigate their inner regions. However, optical and X-ray results show many discrepancies that have not been fully understood yet. The aim of this paper is to study the "synapses" between the X-ray and optical classifications. For the first time, the new EFLUXER task allowed us to analyse broad band X-ray spectra of emission line nuclei (ELN) without any prior spectral fitting using artificial neural networks (ANNs). Our sample comprises 162 XMM-Newton/pn spectra of 90 local ELN in the Palomar sample. It includes starbursts (SB), transition objects (T2), LINERs (L1.8 and L2), and Seyferts (S1, S1.8, and S2). The ANNs are 90% efficient at classifying the trained classes S1, S1.8, and SB. The S1 and S1.8 classes show a wide range of S1- and S1.8-like components. We suggest that this is related to a large degree of obscuration at X-rays. The S1, S1.8, S2, L1.8, L2/T2/SB-AGN (SB with indications of AGN), and SB classes have similar average X-ray spectra within each class, but these average spectra can be distinguished from class to class. The S2 (L1.8) class is linked to the S1.8 (S1) class with larger SB-like component than the S1.8 (S1) class. The L2, T2, and SB-AGN classes conform a class in the X-rays similar to the S2 class albeit with larger fractions of SB-like component. This SB-like component is the contribution of the star-formation in the host galaxy, which is large when the AGN is weak. An AGN-like component seems to be present in the vast majority of the ELN, attending to the non-negligible fraction of S1-like or S1.8-like component. This trained ANN could be used to infer optical properties from X-ray spectra in surveys like eRosita.