Local Outlier Detection Algorithm Based on Multi-Label Learning

The density-based local outlier detection algorithm (LOF) is not suitable for detecting the data set which is high dimension and polysemous. In this paper, we popose a local outlier detection algorithm based on multi-label learing(MLL-LOF).The main ideal of the MLL-LOF algorithm is as follows: Firstly, the real object data is divided into multi-instance by using an MLL framework, then the MLL-LOF calculates the final outlier factor and detects outliers by using degradation strategy and weight adjustment. We compare the MLL-LOF algorithm with other classical local outlier detection algorithms by using actual data set which is comes from enterprise monitoring. Experimental results show that the precision, recall, F1 and time efficiency of the MLL-LOF algorithm are superior to the traditional local outlier detection algorithm.