A Class Of Non Parametric Method For Software Failure Prediction

The prediction of software failure in software reliability is the most time-consuming and difficult part in the whole software formation process, and any improvement in failure prediction can greatly reduce the cost of software. The traditional software failure prediction method was based on the classical parameter model, which could not only estimate the prediction model hardly, but also the error rate was high, and the efficiency was low. Therefore, the use of non-parametric theory and technology, so that in the process of software failure prediction does not depend on the empirical assumptions and the estimation of the unknown model, which avoids the problem of estimating the deviation of the model parameters, and can make full use of the historical information of the software failure data. Based on the non-parametric method, the software prediction is based on the historical information of the software failure data. Then, the model is used to predict the failure of the existing data. On the basis of analyzing the influence factors of software failure, this paper studied and improved a kind of traditional non-parametric model based on the improvement of the accuracy of software failure prediction. The main work was as follows:First, the single variable and non-parametric software failure prediction method of weighted sliding windows was proposed(WSW-NM), and the software failure data of the single variable was predicted. Using the sliding window as input, the impact of the recent data on the software failure number was increased by setting the exponential weight function, and the data in the sliding window was weighted to deal with the abnormal points. Experiments showed that the proposed method could improve the prediction accuracy of software failure.Second, based on principal component analysis(PCA), a multivariable nonparametric software failure prediction method(PCA-INW) was proposed. Principal component of the raw data reduced the input factor number of sample analysis and variance contribution rate of the principal component analysis as the weight of kernel bandwidth matrix of the non-parametric method, eliminated the effects of each input factor on the effects caused by difference. The experimental results showed that the training time and the prediction time were reduced, and the accuracy and stability of the prediction results were improved.Third, multivariable non-parametric software failure prediction based on non-parametric causality test, for the Granger causality test method was only applicable to two variables and the limit of data stability, a new method based on improved nonparametric estimation method(PCA-INW) was proposed(PCA-INW-Granger). The method using the PCA-INW, from the relationship between complex system modeling theory research variables, put forward the applicable in examining multi-dimensional variable causality test method based on Granger causality and its advantages was for single output(result) multidimensional system, without many restrictions, could test in the nonlinear causality between variables.Finally, in order to verify the effectiveness of the proposed method, this paper used a set of standard data set as the research object, and the results showed that the proposed method had better adaptability than the traditional method, and the accuracy and stability of the proposed method were significantly improved.