Why annotation results are totally not correct

[l-magnificence]

I used GSE115600 for example, which were LogNormalized before runing AzimuthAPI.
However, the annotation results were totally not correct. All the neural cells were annotated as immune cells.

table(sc$Cell_type,sc$full_hierarchical_labels)
                                 Immune cell|Lymphoid cell|T/NK cell|T cell|CD4 T cell|Naive CD4 T cell
  aNSC1                                                                    609
  aNSC0                                                                    452
  qNSC1                                                                    450
  aNSC2                                                                    382
  qNSC2                                                                    315
  NB                                                                       298
  TAP                                                                      435
  OD                                                                        83
  OPC                                                                       44

table(sc$azimuth_broad)
      False Immune cell 
       3052          16

I wonder how can this package be accurately applied for our own data?