Method and equipment for determining abnormal reasons of multi-dimensional sample data

Abstract

The invention provides a method for determining an abnormality reason of multi-dimensional sample data, which comprises the steps of obtaining the multi-dimensional sample data, generating an isolated forest model, inputting the multi-dimensional sample data into the isolated forest model to detect abnormality, and recording a final dimension and a corresponding statistical frequency, and determining the final occurrence probability of the final dimension and evaluating the contribution level of the final dimension to the exception. An apparatus for determining causes of anomalies of multi-dimensional sample data and a computer-readable storage medium are also presented. According to the method and the equipment, automatic anomaly reason analysis after anomaly value detection of the multi-dimensional sample data in complex distribution can be completed.

Description

technical field [0001] The present application relates to risk control, in particular to a method and equipment for determining abnormal causes of multi-dimensional sample data applied in the catering industry. Background technique [0002] In an industry such as the restaurant industry, there is a need for risk control of the transactional data of the people involved. The premise of risk control is to accurately detect abnormal behaviors in the transaction data of restaurant personnel. [0003] The transaction data of restaurant personnel usually has the characteristics of high dimensionality, multi-peak and complex joint distribution. The abnormal sample data in the sample data set is different from most sample data, and it accounts for a small proportion of the overall data. Therefore, traditional classification methods based on supervised learning, such as classification algorithms such as SVM and logistic regression, which use a large number of positive sample data and...

AI Technical Summary

Problems solved by technology

However, although the single-dimensional outlier detection algorithm has good interpretability, it can only deal with sample data with a good distribution shape and cannot include the joint distribution characteristics between each dimension into the evaluation range, so the correlation between each dimension cannot be evaluated. feature

Although the high-dimensional outlier detection algorithm represented by the isolated forest model algorithm can well detect the complex joint distribution of high-dimensional sample data, the current mainstream application scheme only provides anomaly scores representing the degree of comprehensive anomaly

The anomaly score cannot represent the abnormal causes of multi-dimensional outliers, and cannot identify the main dimensions that are abnormal at the outliers and their proportions to the abnormal causes, that is, the abnormal detection results are not interpretable

Images