Research On Software Defect Prediction Based On Extreme Learning Machine

Software Defect Prediction(SDP)is currently one of the research fields that domestic and foreign experts and scholars are more interested in.With limited resources,it is difficult to find all defective modules through traditional software testing methods.Many researchers use machine learning methods to build SDP models to determine whether software modules are defective.Current research on software defect prediction is also faced with the problem of insufficient labeled defective modules in the past.Generally,in order to alleviate this problem,some researchers will use semi-supervised learning to make full use of unlabeled modules to enhance the training effect.Although the existing semi-supervised defect prediction methods have achieved good results,there is still much room for improvement.Extreme Learning Machine(ELM)is a feedforward neural network with only one hidden layer.It can randomly specify the weight value and bias vector of the input layer,and use the generalized inverse matrix theory to solve the output layer at one time.To complete the construction of the predictive model.Extreme learning machine has the advantages of simple structure,strong generalization and fast training speed.This article will conduct research on software defect prediction based on extreme learning machines.The content is briefly summarized as follows:First,an improved SMOTE based multi-kernel extreme learning machine(SMMKELM)is proposed.This method first uses the k-means&SMOTE method to process the class imbalance of the data to alleviate the problem of the data class imbalance,and then uses the stacked denoising autoencoder to perform deep feature extraction on the processed data to make up for the lack of feature learning of the extreme learning machine.Base on the extreme learning machine,the kernel method is introduced to improve the extreme learning machine that is not ideal for feature learning,so that the original model can further learn the module features,and improving the classification accuracy.Secondly,in order to deal with the large number of unmarked modules in the practical application of software defect prediction research,and to make full use of these unmarked modules to make the actual operation more feasible,this article proposes a semi-supervised extreme learning machine based on improved SMOTE using for software defect prediction(SELM).In order to alleviate the class imbalance problem of the sample,this article uses k-means&SMOTE to deal with the class imbalance of the modules.This method first uses k-means&SMOTE to process the class imbalance of the modules,then uses a stacked denoising autoencoder for feature extraction,and finally uses a semi-supervised extreme learning machine for classification.Finally,in order to alleviate the negative effects of some unlabeled modules on the classification work,this article proposes a software defect prediction method based on the improved SMOTE adaptive safe semi-supervised extreme learning machine(ASELM).This method is mainly based on the SELM method and introduces the concept of safety degree to the use of unlabeled modules for semi-supervised extreme learning machines,and reduces the negative effects of some unlabeled modules by marking the safety degree of unlabeled modules.Using this method can further improve the utilization efficiency of unlabeled modules,so that the classification effect of the semi-supervised extreme learning machine is improved.