With the Long-Term Evolution (LTE) of mobile networks, the types of network fault have become more complex and diverse. In order to ensure reliable and safe run of the network, it is increasingly difficult for traditional manual administration and maintenance to cope with the complicated and heavy network faults. The super machine learning ability of artificial intelligence can well realize intelligent prediction and processing of network faults by sorting and analyzing a large amount of Key Performance Indicator (KPI) data on network operation and maintenance. In the face of complex network fault types, the number of labeled KPI samples used for training the model is very limited and it is difficult to obtain the samples. This paper proposes a multiclass classification algorithm based on semi-supervised transductive support vector machine for fault root cause detection in LTE networks, and compares its performance with supervised learning and unsupervised learning algorithms. Theoretical analysis and achieved results show that the algorithm proposed in this paper achieves a good learning effect on the mixed sample training set composed of a small number of labeled samples and a large number of unlabeled samples, and has a high classification accuracy for unlabeled samples.