Top (k1, k2) Distance-based outliers detection in an uncertain dataset

In this paper, we focus on distance-based outliers detection in an uncertain dataset, which is very useful in large social network. Based on the x-tuple model and the possible world semantics, we propose the concept of tuple outlier score, top k₁ probability and top (k₁, k₂) distance-based outlier. We then design an algorithm using dynamic programming technique to calculate tuple outlier scores and detect top (k₁, k₂) distance-based outliers. The local neighbor region is proposed to detect approximate outliers with high precision efficiently. We also propose two pruning strategies to avoid additional computation overhead and prune data objects that cannot be outliers. After theory analysis, we conduct experiments in two real datasets to verify good performance of our method.