A system fault detection method based on multi-source class imbalance data

Abstract

The application relates to the technical field of machine learning and data mining, and particularly relates to a system fault detection method based on multi-source class imbalance data, which comprises the following steps: acquiring multi-source time series data of each server in a system, and obtaining an initial data set according to the multi-source time series data; performing feature engineering on the initial data set to obtain a structured data set; processing the structured data set by using a label-specific distance measurement algorithm to obtain an instance set and a near neighbor instance set; dividing the structured data set by using a multi-label hierarchical sampling method to obtain a training set and a test set; sampling the instance set and the near neighbor instance set to obtain a sampled training set; and establishing a classifier chain model, inputting the sampled training set into the classifier chain model for training to obtain a trained classifier chain model; and the label-specific sampling method can effectively balance the instance numbers of various fault sources, and can ensure that each fault class can be fully paid attention to in the training process.

Images