The purpose of outlier detection is to identify data points that are significantly different or inconsistent with other data in a given dataset. Due to its applications in various fields such as network intrusion detection, fraud detection, and life sciences, outlier detection has become a research hotspot in the field of data mining. This paper takes the non-stationary multi-parameter dataset from a distributed soft-sensing module as the research object. By combining the fast density peak clustering algorithm and the natural neighbor search algorithm, an improved density peak clustering outlier detection algorithm is proposed for experimental study. In the proposed global-local outlier decision graph outlier explanation method, global outliers appear in the upper right corner of the decision graph, while local outliers are sparsely distributed in the upper middle part of the decision graph, allowing the density distribution of normal data points in each cluster to be observed. Additionally, experiments on artificial and real datasets demonstrate that the IDPCOD algorithm can efficiently and comprehensively detect outliers in nonstationary multi-parameter data.
