Defect Prediction Performance Research Articles

An Empirical Study of the Impact of Class Overlap on the Performance and Interpretability of Cross-Version Defect Prediction

The class overlap problem refers to instances from different categories heavily overlapping in the feature space. This issue is one of the challenges in improving the performance of software defect prediction (SDP). Currently, the studies on the impact of class overlap on SDP mainly focused on within-project defect prediction and cross-project defect prediction. Moreover, the existing class overlap instances cleaning methods are not suitable for cross-version defect prediction. In this paper, we propose a class overlap instances cleaning method based on the Ratio of K-nearest neighbors with the Same Label (RKSL). This method removes instances with the abnormal neighbor ratio in the training set. Based on the RKSL method, we investigate the impact of class overlap on the performance and interpretability of the cross-version defect prediction model. The experiment results show that class overlap can affect the performance of cross-version defect prediction models significantly. The RKSL method can handle the class overlap problem in defect datasets, but it may impact the interpretability of models. Through the analysis of feature changes, we consider that class overlap instances cleaning can assist models in identifying more important features.

Semantic and traditional feature fusion for software defect prediction using hybrid deep learning model

Software defect prediction aims to find a reliable method for predicting defects in a particular software project and assisting software engineers in allocating limited resources to release high-quality software products. While most earlier research has concentrated on employing traditional features, current methodologies are increasingly directed toward extracting semantic features from source code. Traditional features often fall short in identifying semantic differences within programs, differences that are essential for the development of reliable and effective prediction models. In contrast, semantic features cannot present statistical metrics about the source code, such as the code size and complexity. Thus, using only one kind of feature negatively affects prediction performance. To bridge the gap between the traditional and semantic features, we propose a novel defect prediction model that integrates traditional and semantic features using a hybrid deep learning approach to address this limitation. Specifically, our model employs a hybrid CNN-MLP classifier: the convolutional neural network (CNN) processes semantic features extracted from projects’ abstract syntax trees (ASTs) using Word2vec. In contrast, the traditional features extracted from the dataset repository are processed by a multilayer perceptron (MLP). Outputs of CNN and MLP are then integrated and fed into a fully connected layer for defect prediction. Extensive experiments are conducted on various open-source projects to validate CNN-MLP’s effectiveness. Experimental results indicate that CNN-MLP can significantly enhance defect prediction performance. Furthermore, CNN-MLP’s improvements outperform existing methods in non-effort-aware and effort-aware cases.

Optimizing Software Defect Prediction Models: Integrating Hybrid Grey Wolf and Particle Swarm Optimization for Enhanced Feature Selection with Popular Gradient Boosting Algorithm

Software defects, also referred to as software bugs, are anomalies or flaws in computer program that cause software to behave unexpectedly or produce incorrect results. These defects can manifest in various forms, including coding errors, design flaws, and logic mistakes, this defect have the potential to emerge at any stage of the software development lifecycle. Traditional prediction models usually have lower prediction performance. To address this issue, this paper proposes a novel prediction model using Hybrid Grey Wolf Optimizer and Particle Swarm Optimization (HGWOPSO). This research aims to determine whether the Hybrid Grey Wolf and Particle Swarm Optimization model could potentially improve the effectiveness of software defect prediction compared to base PSO and GWO algorithms without hybridization. Furthermore, this study aims to determine the effectiveness of different Gradient Boosting Algorithm classification algorithms when combined with HGWOPSO feature selection in predicting software defects. The study utilizes 13 NASA MDP dataset. These dataset are divided into testing and training data using 10-fold cross-validation. After data is divided, SMOTE technique is employed in training data. This technique generates synthetic samples to balance the dataset, ensuring better performance of the predictive model. Subsequently feature selection is conducted using HGWOPSO Algorithm. Each subset of the NASA MDP dataset will be processed by three boosting classification algorithms namely XGBoost, LightGBM, and CatBoost. Performance evaluation is based on the Area under the ROC Curve (AUC) value. Average AUC values yielded by HGWOPSO XGBoost, HGWOPSO LightGBM, and HGWOPSO CatBoost are 0.891, 0.881, and 0.894, respectively. Results of this study indicated that utilizing the HGWOPSO algorithm improved AUC performance compared to the base GWO and PSO algorithms. Specifically, HGWOPSO CatBoost achieved the highest AUC of 0.894. This represents a 6.5% increase in AUC with a significance value of 0.00552 compared to PSO CatBoost, and a 6.3% AUC increase with a significance value of 0.00148 compared to GWO CatBoost. This study demonstrated that HGWOPSO significantly improves the performance of software defect prediction. The implication of this research is to enhance software defect prediction models by incorporating hybrid optimization techniques and combining them with gradient boosting algorithms, which can potentially identify and address defects more accurately

Comparative Study of Various Hyperparameter Tuning on Random Forest Classification With SMOTE and Feature Selection Using Genetic Algorithm in Software Defect Prediction

Comparative Study of Various Hyperparameter Tuning on Random Forest Classification With SMOTE and Feature Selection Using Genetic Algorithm in Software Defect Prediction

Efficient Cross-Project Software Defect Prediction Based on Federated Meta-Learning

Software defect prediction is an important part of software development, which aims to use existing historical data to predict future software defects. Focusing on the model performance and communication efficiency of cross-project software defect prediction, this paper proposes an efficient communication-based federated meta-learning (ECFML) algorithm. The lightweight MobileViT network is used as the meta-learner of the Model Agnostic Meta-Learning (MAML) algorithm. By learning common knowledge on the local data of multiple clients, and then fine-tuning the model, the number of unnecessary iterations is reduced, and communication efficiency is improved while reducing the number of parameters. The gradient information model is encrypted using the differential privacy of the Laplace mechanism, and the optimal privacy budget is determined through experiments. Experiments on three public datasets (AEEEM, NASA, and Relink) verified the effectiveness of ECFML in terms of parameter quantity, convergence, and model performance of cross-project software defect prediction.

Comparative analysis of the performance of supervised learning algorithms for photovoltaic system fault diagnosis

New trends were introduced in using PhotoVoltaic (PV) energy which are mostly attributable to new laws internationally having a goal to decrease the usage of fossil fuels. The PV systems efficiency is impacted significantly by environmental factors and different faults occurrence. These faults if they were not rapidly identified and fixed may cause dangerous consequences. A lot of methods have been introduced in the literature to detect faults that may occur in a PV system such as using Current-Voltage (I-V) curve measurements, atmospheric models and statistical methods. In this paper, various machine learning techniques in particular supervised learning techniques are used for PV array failure diagnosis. The main target is the identification and categorization of several faults that may occur such as shadowing, degradation, open circuit and short circuit faults that have a great impact on PV systems performance. The results showed the technique’s high ability of fault diagnosis capability. The K-Nearest Neighbor (KNN) technique showed the best fault prediction performance. It achieves prediction accuracy of 99.2% and 99.7% Area Under Curve-Receiver Operating Curve (AUC-ROC) score. This shows its superiority in fault prediction in PV systems over other used methods Decision Tree, Naïve Bayes, and Logistic Regression.

Evaluating the Impact of Data Transformation Techniques on the Performance and Interpretability of Software Defect Prediction Models

The performance of software defect prediction (SDP) models determines the priority of test resource allocation. Researchers also use interpretability techniques to gain empirical knowledge about software quality from SDP models. However, SDP methods designed in the past research rarely consider the impact of data transformation methods, simple but commonly used preprocessing techniques, on the performance and interpretability of SDP models. Therefore, in this paper, we investigate the impact of three data transformation methods (Log, Minmax, and Z-score) on the performance and interpretability of SDP models. Through empirical research on (i) six classification techniques (random forest, decision tree, logistic regression, Naive Bayes, K-nearest neighbors, and multilayer perceptron), (ii) six performance evaluation indicators (Accuracy, Precision, Recall, F1, MCC, and AUC), (iii) two interpretable methods (permutation and SHAP), (iv) two feature importance measures (Top-k feature rank overlap and difference), and (v) three datasets (Promise, Relink, and AEEEM), our results show that the data transformation methods can significantly improve the performance of the SDP models and greatly affect the variation of the most important features. Specifically, the impact of data transformation methods on the performance and interpretability of SDP models depends on the classification techniques and evaluation indicators. We observe that log transformation improves NB model performance by 7%–61% on the other five indicators with a 5% drop in Precision. Minmax and Z-score transformation improves NB model performance by 2%–9% across all indicators. However, all three transformation methods lead to substantial changes in the Top-5 important feature ranks, with differences exceeding 2 in 40%–80% of cases (detailed results available in the main content). Based on our findings, we recommend that (1) considering the impact of data transformation methods on model performance and interpretability when designing SDP approaches as transformations can improve model accuracy, and potentially obscure important features, which lead to challenges in interpretation, (2) conducting comparative experiments with and without the transformations to validate the effectiveness of proposed methods which are designed to improve the prediction performance, and (3) tracking changes in the most important features before and after applying data transformation methods to ensure precise and traceable interpretability conclusions to gain insights. Our study reminds researchers and practitioners of the need for comprehensive considerations even when using other similar simple data processing methods.

Mutation‐based data augmentation for software defect prediction

AbstractSoftware defect prediction (SDP) aims to distinguish between defective and nondefective instances, but the imbalance between these two classes often leads to reduced prediction performance. Conventional SDP approaches use oversampling techniques, such as synthetic oversampling, to tackle the problem of imbalanced data. However, these methods merely synthesize new instances based on traditional code features without considering actual defects at the code level. To address the issue of data imbalance while preserving semantic features of code samples, a mutation‐based data augmentation approach in SDP is proposed. The method utilizes the mutation operator to generate mutants that mutate nondefective instances and create new defective instances. Six projects from the PROMISE dataset are used to evaluate the approach, employing four traditional and two deep classifiers. The experimental results demonstrate the effectiveness of this method in improving defect prediction performance for both traditional and deep classifiers compared with other data augmentation methods.

Software defect prediction using a bidirectional LSTM network combined with oversampling techniques

AbstractSoftware defects are a critical issue in software development that can lead to system failures and cause significant financial losses. Predicting software defects is a vital aspect of ensuring software quality. This can significantly impact both saving time and reducing the overall cost of software testing. During the software defect prediction (SDP) process, automated tools attempt to predict defects in the source codes based on software metrics. Several SDP models have been proposed to identify and prevent defects before they occur. In recent years, recurrent neural network (RNN) techniques have gained attention for their ability to handle sequential data and learn complex patterns. Still, these techniques are not always suitable for predicting software defects due to the problem of imbalanced data. To deal with this problem, this study aims to combine a bidirectional long short-term memory (Bi-LSTM) network with oversampling techniques. To establish the effectiveness and efficiency of the proposed model, the experiments have been conducted on benchmark datasets obtained from the PROMISE repository. The experimental results have been compared and evaluated in terms of accuracy, precision, recall, f-measure, Matthew’s correlation coefficient (MCC), the area under the ROC curve (AUC), the area under the precision-recall curve (AUCPR) and mean square error (MSE). The average accuracy of the proposed model on the original and balanced datasets (using random oversampling and SMOTE) was 88%, 94%, And 92%, respectively. The results showed that the proposed Bi-LSTM on the balanced datasets (using random oversampling and SMOTE) improves the average accuracy by 6 and 4% compared to the original datasets. The average F-measure of the proposed model on the original and balanced datasets (using random oversampling and SMOTE) were 51%, 94%, And 92%, respectively. The results showed that the proposed Bi-LSTM on the balanced datasets (using random oversampling and SMOTE) improves the average F-measure by 43 and 41% compared to the original datasets. The experimental results demonstrated that combining the Bi-LSTM network with oversampling techniques positively affects defect prediction performance in datasets with imbalanced class distributions.

A Novel Developed Supervised Machine Learning System For Classification And Prediction of Software Faults Using NASA Dataset

The software systems of modern computers are extremely complex and versatile. Therefore, it is essential to regularly detect and correct software design faults. In order to devote resources effectively towards the creation of trustworthy software, software companies are increasingly engaging in the practise of predicting fault-prone modules in advance of testing. These software fault prediction methods rely on the thoroughness with which prior software versions' fault as well as related code has been retrievedTime, energy, and money are all saved as a result. Increases the company's initial success and bottom line greatly by satisfying its clientele. Numerous academics have poured into this area throughout the years in an effort to raise the bar for all software. Nowadays, The most often used approaches in this field are those based on machine learning (ML). The field of ML seeks to perfect software capable of evolving as well as adapting in response to fresh data. This paper introduces a fresh approach for doing ML by bringing together a number of different expert systems. In order to reach agreement on which aspects of a software system need to be tested, the proposed multi-classifier model pools the strengths of the most effective classifiers. Several top-performing classifiers for defect prediction are put through their paces in an experiential evaluation. We test our method on 16 publicly available datasets from the NASA Metric Data Programme (MDP) repository at the promise repository. Parameters of confusion, recall, precision, recognition accuracy, etc., are evaluated and contrasted with existing schemes in a software analysis performed with the help of the python simulation tool with findings. The experimental outcomes demonstrate that by combining LGBM, XGBoost, and Voting classifiers, using a multi classifier approach, we are capable to significantly improve software fault prediction performance. The results of the investigation show that the suggested method will lead to better practical outcomes in the prediction of device failures.

Towards an understanding of intra-defect associations: Implications for defect prediction

In previous studies, when collecting defect data, if the fix of a defect spans multiple modules, each involved module is labeled as defective. In this context, the defect prediction models are built based on the features of each individual module, ignoring the potential associations between the modules involved in the same defect(referred to as “intra-defect associations”). Considering the possibility of numerous cross-module defects in practice, we hypothesize that these intra-defect associations could play a crucial role in enhancing defect prediction performance. Unfortunately, there is no empirical evidence to know that. To this end, we are motivated to conduct a comprehensive study to explore the implications of intra-defect associations for defect prediction. We first examine the proportion of cross-module defects and the relationships between the involved modules. The results reveal that, at function level, the majority of defects occur across functions, with most of the cross-module defects exhibiting implicit dependencies. Inspired by these findings, we propose a novel data processing approach for building defect prediction models. This approach leverages the intra-defect associations by merging the involved modules into new instances with mean or median variables to augment the training data. The experimental results indicate that considering intra-defect associations can significantly improve the defect prediction performance in both the ranking and classification scenarios. This study provides valuable insights into the implications of intra-defect associations for defect prediction.

Feature Selection Using Firefly Algorithm With Tree-Based Classification In Software Defect Prediction

Defects that occur in software products are a universal occurrence. Software defect prediction is usually carried out to determine the performance, accuracy, precision and performance of the prediction model or method used in research, using various kinds of datasets. Software defect prediction is one of the Software Engineering studies that is of great concern to researchers. This research was conducted to determine the performance of tree-based classification algorithms including Decision Trees, Random Forests and Deep Forests without using feature selection and using firefly feature selection. And also know the tree-based classification algorithm with firefly feature selection which can provide better software defect prediction performance. The dataset used in this study is the ReLink dataset which consists of Apache, Safe and Zxing. Then the data is divided into testing data and training data with 10-fold cross validation. Then feature selection is performed using the Firefly Algorithm. Each ReLink dataset will be processed by each tree-based classification algorithm, namely Decision Tree, Random Forest and Deep Forest according to the results of the firefly feature selection. Performance evaluation uses the AUC value (Area under the ROC Curve). Research was conducted using google collab and the average AUC value generated by Firefly-Decision Tree is 0.66, the average AUC value generated by Firefly-Random Forest is 0.77, and the average AUC value generated by Firefly-Deep Forest is 0, 76. The results of this study indicate that the approach using the Firefly algorithm with Random Forest classification can work better in predicting software damage compared to other tree-based algorithms. In previous studies, tree-based classification with hyperparameter tuning on software defect prediction datasets obtained quite good results. In another study, the classification performance of SVM, Naïve Bayes and K-nearest neighbor with firefly feature selection resulted in improved performance. Therefore, this research was conducted to determine the performance of a tree-based algorithm using the firefly selection feature.

A Software Defect Prediction Approach Based on Hybrid Feature Dimensionality Reduction

Software defect prediction (SDP) is designed to assist software testing, which can reasonably allocate test resources to reduce costs and improve development efficiency. In order to improve the prediction performance, researchers have designed many defect-related features for SDP. However, feature redundancy (FR) and irrelevance caused by the increasing dimensions of data will greatly degrade the performance of defect prediction. In order to solve the problems, researchers have proposed various data dimensionality reduction methods. These methods can be simply divided into two categories of methods: feature selection and feature extraction. However, the two categories of methods have their own advantages and limitation. In this paper, we propose a Hybrid Feature Dimensionality Reduction Approach (HFDRA) for SDP, which combines the two different kinds of methods, to improve the performance of SDP. HFDRA approach can be divided into two stages: feature selection and feature extraction. First, HFDRA divides the original features into several feature subsets through a clustering algorithm in the feature selection stage. Then, in the feature extraction stage, kernel principal component analysis (KPCA) is used to reduce the dimensionality of each feature subset. Finally, the reduced-dimensional data is used to build the prediction model. In the empirical study, we use 22 projects from AEEEM, SOFTLAB, MORP, and ReLink as experiment object. In this paper, we first compare our approach with seven baseline methods and three state-of-the-art methods. Then, we analyze the relationship between FR and prediction performance. Experiment results show that our approach outperforms the state-of-the-art data dimensionality reduction methods for defect prediction.

Hybrid Defect Prediction Model Based on Counterfactual Feature Optimization

Software defect prediction is critical to ensuring software quality. Researchers have worked on building various defect prediction models to improve the performance of defect prediction. Existing defect prediction models are mainly divided into two categories: models constructed based on artificial statistical features and models constructed based on semantic features. DP-CNN [Li J, He P, Zhu J, et al. Software defect prediction via convolutional neural network. In: 2017 IEEE international conference on software quality, reliability and security (QRS). IEEE, 2017; 318–328.] is one of the best defect prediction models, because it combines both artificial statistical features and semantic features, so its performance is greatly improved compared to traditional defect prediction models. This paper is based on the DP-CNN model and makes the following two improvements: first, using a new Struc2vec network representation technique to mine existing information between software modules, which specializes in learning node representations from structural identity and can further extract structural features associated with defects. Let the DP-CNN model once again incorporate the newly mined structural features. Then, this paper proposes a feature selection method based on counterfactual explanations, which can determine the importance score of each feature by the feature change rate of counterfactual samples. The origin of these feature importance scores is interpretable. Under the guidance of these interpretable feature importance scores, better feature subsets can be obtained and used to optimize artificial statistical features within the DP-CNN model. Based on the above methods, this paper proposes a new hybrid defect prediction model DPS-CNN-STR. Evaluating our model on six open source projects in terms of F1 score in defect prediction. Experimental results show that DPS-CNN-STR improves the state-of-the-art method by an average of 3.3%.

A comprehensive diagnosis method of valve leakage faults based on bi-sensor information fusion

Acoustic emission (AE) signals caused by valve leakage exhibit obvious nonlinearity and nonstationarity characteristics. Due to the limitations of traditional valve leakage diagnosis methods, it is difficult to distinguish between internal and external valve leakage failures effectively. Recognizing this challenge, a comprehensive valve leakage diagnosis method based on a multichannel fusion convolutional neural network (MCFCNN) is proposed. First, AE signals are converted from one-dimensional time-domain signals to two-dimensional time–frequency images by the time–frequency analysis method. Then, the time–frequency images are used as model inputs, and MCFCNN fuses the features of time–frequency images from two different position. Hence, a new comprehensive diagnosis method for the bi-sensor fusion contains the time–frequency information, modal information, and position information of valve leakage is proposed. Subsequently, the effectiveness of the proposed method was verified through valve leakage simulation experiments. Furthermore, in order to study the impact of modal information on identifying internal and external valve leakage faults, the fault prediction performance of MCFCNN was compared and analyzed using short-time Fourier transform (STFT) and time-reassigned synchrosqueezing transform (TSST). Finally, according to the needs of engineering practice, the impact of sampling length on different methods is studied. The results show that compared to STFT–MCFCNN, TSST–MCFCNN required a shorter sampling length with the same diagnostic accuracy, which means that the method proposed in this study can achieve faster response time for ball valve leakage under conventional leakage flow rates.

Cross‐version defect prediction using threshold‐based active learning

AbstractBecause defects in software modules (e.g., classes) might lead to product failure and financial loss, software defect prediction enables us to better understand and control software quality. Software development is a dynamic evolutionary process that may result in data distributions (e.g., defect characteristics) varying from version to version. In this case, effective cross‐version defect prediction (CVDP) is not easy to achieve. In this paper, we aim to investigate whether the defect prediction method of the threshold‐based active learning (TAL) can tackle the problem of the different data distribution between successive versions. Our TAL method includes two stages. At the active learning stage, a committee of investigated metrics is constructed to vote on the unlabeled modules of the current version. We pick up the unlabeled module with the median of voting scores to domain experts. The domain experts test and label the selected unlabeled module. Then, we merge the selected labeled module and the remaining modules with pseudo‐labels from the current version into the labeled modules of the prior version to form enhanced training data. Based on the training data, we derive the metric thresholds used for the next iteration. At the defect prediction stage, the iterations stop when a predefined threshold is reached. Finally, we use the cutoff threshold of voting scores, that is, 50%, to predict the defect‐prone of the remaining unlabeled modules. We evaluate the TAL method on 31 versions of 10 projects with three prevalent performance indicators. The results show that TAL outperforms the baseline methods, including three variations methods, two common supervised methods, and the state‐of‐the‐art method Hybrid Active Learning and Kernel PCA (HALKP). The results indicate that TAL can effectively address the different data distribution between successive versions. Furthermore, to keep the cost of extensive testing low in practice, selecting 5% of candidate modules from the current version is sufficient for TAL to achieve a good performance of defect prediction.

Ensemble Machine Learning Paradigms in Software Defect Prediction

Predicting faults in software aims to detect defects before the testing phase, allowing for better resource allocation and high-quality software development, which is a requisite for any organization. Machine learning techniques aid in the resolution of such issues and a variety of predictive models are being developed to categorize the software into defective modules and the one which is non-defective ones. Though applying these advanced machine learning techniques results in better utilization of time and other resources, there is still poor prediction as reported in many studies. This is because of several challenges that block defective software data, including redundancy, correlation, feature irrelevance, missing samples, and an imbalanced distribution between the faulty and non-faulty classes. Ensemble Machine learning has been adopted by practitioners and researchers globally to deal with such problems, and it is proven to demonstrate some improvement in defect prediction performance. In this review paper, all ensemble-based machine learning techniques developed for software defect prediction from 2018 to 2021 have been critically analyzed. The nucleus of this paper is to get a deep insight into why the various hybrid models still suffer from poor performance on the available datasets. A detailed review with a focus on multiple perspectives viz. faulty and non-defective datasets, performance evaluation criteria, and machine learning techniques have revealed certain gaps that can be addressed by developing more robust hyperparameter optimization algorithms, feature engineering, developing stacking and averaging models.

Improving Cross-Project Software Defect Prediction Method Through Transformation and Feature Selection Approach

In the traditional software defect prediction methodology, the historical record (dataset) of the same project is partitioned into training and testing data. In a practical situation where the project to be predicted is new, traditional software defect prediction cannot be employed. An alternative method is cross-project defect prediction, where the historical record of one project (source) is used to predict the defect status of another project (target). The cross-project defect prediction method solves the limitations of the historical records in the traditional software defect prediction method. However, the performance of cross-project defect prediction is relatively low because of the distribution differences between the source and target projects. Furthermore, the software defect dataset used for cross-project defect prediction is characterized by high-dimensional features, some of which are irrelevant and contribute to low performance. To resolve these two issues, this study proposes a transformation and feature selection approach to reduce the distribution difference and high-dimensional features in cross-project defect prediction. A comparative experiment was conducted on publicly available datasets from the AEEEM. Analysis of the results obtained shows that the proposed approach in conjugation with random forest as the classification model outperformed the other four state-of-the-art cross-project defect prediction methods based on the commonly used performance evaluation metric F1_score.

Software defect prediction with semantic and structural information of codes based on Graph Neural Networks

Most defect prediction methods consider a series of traditional manually designed static code metrics. However, only using these hand-crafted features is impractical. Some researchers use the Convolutional Neural Network (CNN) to capture the potential semantic information based on the program’s Syntax Trees (ASTs). In recent years, leveraging the dependency relationships between software modules to construct a software network and using network embedding models to capture the structural information have been helpful in defect prediction. This paper simultaneously takes the semantic and structural information into account and proposes a method called CGCN. This study aims to validate the feasibility and performance of the proposed method in software defect prediction. Abstract Syntax Trees and a Class Dependency Network (CDN) are first generated based on the source code. For ASTs, symbolic tokens are extracted and encoded into vectors. The numerical vectors are then used as input to the CNN to capture the semantic information. For CDN, a Graph Convolutional Network (GCN) is used to learn the structural information of the network automatically. Afterward, the learned semantic and structural information are combined with different weights. Finally, we concatenate the learned features with traditional hand-crafted features to train a classifier for more accurate defect prediction. The proposed method outperforms the state-of-the-art defect prediction models for both within-project prediction (including within-version and cross-version) and cross-project prediction on 21 open-source projects. In general, within-version prediction achieves better performance in the three prediction tasks. The proposed method of combining semantic and structural information can improve the performance of software defect prediction. • A novel defect prediction model (CGCN), which combines CNN and GCN. The proposed model can extract the semantic information from the ASTs of the source codes, as well as the structural information between modules from the software network. Therefore, the defect prediction performance is improved. • Different weight parameters are used for semantic and structural features, in order to make them play different contributions in different tasks and improve the model generalization ability. • The efficiency of CGCN is evaluated on seven Java open-source projects (each project selects three versions, with a total of 21 datasets) and three tasks (withinversion defect prediction, cross-version defect prediction and cross-project defect prediction). The results demonstrate that the CGCN variant using weight parameters outperforms the state-of-the-art methods, while the CGCN variant without using weight parameters outperforms the benchmarks only in the withinversion defect prediction. • Different from “ GCN2defect: Graph Convolutional Networks for SMOTETomekbased Software Defect Prediction ” on 2021 IEEE 32nd International Symposium on Software Reliability Engineering (ISSRE), in addition to the features of software network, this paper also combines the semantic features of source codes, and extends to cross-version and cross-project predictions. Experimental results show that the proposed method in this paper has better robustness.

CODE: A Moving-Window-Based Framework for Detecting Concept Drift in Software Defect Prediction

Concept drift (CD) refers to data distributions that may vary after a minimum stable period. CD negatively influences models’ performance of software defect prediction (SDP) trained on past datasets when applied to the new datasets. Based on previous studies of SDP, it is confirmed that the accuracy of prediction models is negatively affected due to changes in data distributions. Moreover, cross-version (CV) defect data are naturally asymmetric due to the nature of their class imbalance. In this paper, a moving window-based concept-drift detection (CODE) framework is proposed to detect CD in chronologically asymmetric defective datasets and to investigate the feasibility of alleviating CD from the data. The proposed CODE framework consists of four steps, in which the first pre-processes the defect datasets and forms CV chronological data, the second constructs the CV defect models, the third calculates the test statistics, and the fourth provides a hypothesis-test-based CD detection method. In prior studies of SDP, it is observed that in an effort to make the data more symmetric, class-rebalancing techniques are utilized, and this improves the prediction performance of the models. The ability of the CODE framework is demonstrated by conducting experiments on 36 versions of 10 software projects. Some of the key findings are: (1) Up to 50% of the chronological-defect datasets are drift-prone while applying the most popular classifiers used from the SDP literature. (2) The class-rebalancing techniques had a positive impact on the prediction performance for CVDP by correctly classifying the CV defective modules and detected CD by up to 31% on the resampled datasets.