Defect prediction method and device based on aerospace software defect data distribution outliers

Abstract

The invention discloses a defect prediction method and device based on aerospace software defect data distribution outliers. The method comprises the steps: constructing an outlier training set and anoutlier test set corresponding to aerospace embedded software defect data according to sample data with an outlier phenomenon; according to the outlier training set, training to obtain a first numberof software defect prediction models; performing test verification on the software defect prediction model according to the outlier test set, and sorting the software defect prediction model according to evaluation indexes to obtain a sorting prediction model; constructing an automatic search optimization algorithm based on a genetic algorithm, and recursively searching replaceable model nodes layer by layer on the basis of a meta-classifier in combination with the sorting prediction model; and according to different evaluation indexes and the model nodes, optimizing to obtain a target modelstructure, and obtaining the defect prediction method of aerospace embedded software defect data distribution outliers. According to the invention, the third-party software evaluation efficiency can be improved.

Description

technical field [0001] The invention relates to the technical field of software defect prediction, in particular to a defect prediction method and device based on outlier points in aerospace software defect data distribution. Background technique [0002] Most of the research on detecting outliers in the data distribution is carried out in the field of statistics, and these studies can be roughly divided into two categories. The first category is based on statistical distribution methods, where a standard distribution (e.g. normal distribution, Poisson distribution, etc.) is used to best fit the data and outliers are defined in terms of probability distributions. The main disadvantage of this type of test is that most of the distributions used are univariate and fitting the data with a standard distribution is expensive and may not yield satisfactory results. The second category is based on spatial depth methods. Represent each data object as a point in space and assign it...

AI Technical Summary

Problems solved by technology

Regarding outlier detection, outliers are more likely to be data objects with a smaller depth, and depth-based methods are inefficient because depth-based methods rely on the calculation of convex hulls and increase complexity

[0003] In the engineering practice of aerospace embedded software defect prediction, because there are outliers in the engineering practice test set in terms of the basic cyclomatic complexity of the measurement element, the number of LCSAJ, the number of non-repeated operators, the depth-first call level, and the number of acyclic paths, etc., Causes poor prediction performance of machine learning classifiers such as conventional Bayesian classifiers, rule classifiers, tree classifiers, meta classifiers, and neural network classifiers

Images