# Why multi-class SVM can't detect some classes?

I use PCA and multi-class SVM for classifying 4-class problem in the Python environment. But in results, I see some differences in detection rate (Unweighted Accuracy in this problem). For example:

• FV | cl.1 % | cl.2 % | cl.3 % | cl.4 % |

• fv1 | 5.4 % | 47.4 % | 2.6 % | 78.6 % |

• fv2 | 2.6 % | 41.6 % | 8.3 % | 78.4 % |

• fv3 | 5.9 % | 49.3 % | 1.8 % | 76.5 % |

you can see that SVM couldn't detect classes 1 and 3 as well as classes 2 and 4. I would be appreciated if anybody helps me to solve this issue.