Software Metrics Research Articles

Software Quality Assurance as a Service: Encompassing the quality assessment of software and services

This paper introduces the Software Quality Assurance as a Service (SQAaaS) concept and it describes an open-source implementation of a comprehensive platform that supports the automated assessment of specific quality metrics for software and services, defined as a set of baseline requirements. The platform is openly accessible, focuses on research software and open science, and promotes best practices by awarding standards-based digital badges to software and services. It provides an easy-to-use web-based graphical user interface which facilitates the interaction with server-side components in charge of automatically creating CI/CD (Continuous Integration/Continuous Delivery) pipelines for automated testing of the baseline criteria. The service is in production and has performed over 2800 assessments, awarding more than 125 digital badges across several scientific disciplines.

Mutation boosted salp swarm optimizer meets rough set theory: A novel approach to software defect detection

AbstractSoftware defect detection (SDD) is crucial to ensure the reliability of software systems and identify defects in classification. One of the key challenges in defect detection is to select more informative and relevant features from the vast pool of available software metrics. A novel approach is proposed in this paper that leverages mutation boosted salp swarm optimizer (MBSSO) and Rough Set Theory for feature selection (FS) in SDD. It efficiently explores the search space and incorporates a mutation boosting mechanism to overcome local optima. Rough Set Theory provides a formal framework for analyzing feature relevance and dependency, while MBSSO optimizes the search for the optimal feature subset. The proposed approach involves encoding the feature subsets using binary representation and designing a fitness function. The MBSSO algorithm explores the feature space, iteratively improving the feature subset based on the fitness function. The selected feature subset is then used to train a defect detection model using the Kernel Extreme Learning Machine (KELM) algorithm. The experimental validation is performed by Project Repository for Software Engineering (PROMISE) dataset and compared the performance of proposed approach against other FS methods and baselines. The experimental validation demonstrates that the proposed approach achieves superior performance and selected feature subset improves the accuracy and efficiency. This research contributes to the advancement of SDD by providing an effective and efficient FS technique using MBSSO and Rough Set Theory.

Evolution of internal dimensions in object‐oriented software–A time series based approach

SummarySoftware evolution is the process of adapting, maintaining, and updating a software system. This process concentrates the most significant part of the software costs. Many works have studied software evolution and found relevant insights, such as Lehman's laws. However, there is a gap in how software systems evolve from an internal dimensions point of view. For instance, the literature has indicated how systems grow, for example, linearly, sub‐linearly, super‐linearly, or following the Pareto distribution. However, a well‐defined pattern of how this phenomenon occurs has not been established. This work aims to define a novel method to analyze and predict software evolution. We based our strategy on time series analysis, linear regression techniques, and trend tests. In this study, we applied the proposed model to investigate how the internal structure of object‐oriented software systems evolves in terms of four dimensions: coupling, inheritance hierarchy, cohesion, and class size. Applying the proposed method, we identify the functions that better explain how the analyzed dimensions evolve. Besides, we investigate how the relationship between dimension metrics behave over the systems' evolution and the set of classes existing in the systems that affect the evolution of these dimensions. We mined and analyzed data from 46 Java‐based open‐source projects. We used eight software metrics regarding the dimensions analyzed in this study. The main results of this study reveal ten software evolution properties, among them: coupling, cohesion, and inheritance evolve linearly; a relevant percentage of classes contributes to coupling and size evolution; a small percentage of classes contributes to cohesion evolution; there is no relation between the software internal dimensions' evolution. The results also indicate that our method can accurately predict how the software system will evolve in short‐term and long‐term predictions.

A proposed model for detecting defects in software projects

<span lang="EN-GB">Defective modules that cause software execution failures are common in large software projects. Source code for a significant number of modules may be found in several software repositories. This software repository includes each module’s software metrics and the module’s faulty status. Software companies face a considerable problem detecting defects in sizeable and complex programming code. In addition, many international reports, such as the comprehensive human appraisal for originating (CHAOS) report, have mentioned that there are countless reasons for the failure of software projects, including the inability to detect errors and defects in the programming code of those projects at an early stage. This research employs a statistical analysis technique to reveal the characteristics that indicate the faulty status of software modules. It is recommended that statistical analysis models derived from the retrieved information be merged with existing project metrics and bug data to improve prediction. When all algorithms are merged with weighted votes, the results indicate enhanced prediction abilities. The proposed statistical analysis outperforms the state-of-the-art method (association rule, decision tree, Naive Bayes, and neural network) in terms of accuracy by 9.1%, 10.3%, 13.1%, and 13.1%, respectively.</span>

Software Testing Using Cuckoo Search Algorithm with Machine Learning Techniques

Software testing are any errors, flaws, bugs, mistakes, failures in a piece of software that might cause the programme to produce incorrect or unexpected results. Testing in software almost always increase both the time and money needed to finish a project. And finding bugs and fixing them is a laborious and expensive software process in and of itself. While it's unrealistic to expect to completely eradicate all testing from a project, their severity may be mitigated. It is possible to predict where bugs may appear in software using a method known as software defect prediction (SDP). The goal of each software development project should be to provide a bug-free product. Predicting where bugs may appear in code, often known as software defect prediction (SDP), is an important part of fixing software. Software of a high calibre should have few bugs. A software metric is a quantitative or qualitative evaluation of some aspect of the programme or its requirements. One of the more recent population-based algorithms, Cuckoo Search (CS) was inspired by the flight patterns of some cuckoo species as well as the Lévy flying patterns of other birds and fruit flies. The needs for international convergence are met by CS. KNN is a significant non-parameter supervised learning technique. This paper presents an overview of Stochastic Diffusion Search (SDS) in the form of a social metaphor to illustrate the processes by which SDS allots resources. The best-fit pattern identification and matching difficulties were addressed by SDS using a novel probabilistic method. As a multiagent population-based global search and optimization method, SDS is a distributed model of computing that makes use of interaction amongst basic agents. The behaviour of SDS is described by studying its resource allocation, convergence to global optimum, resilience, minimum convergence criterion, and linear time complexity within a rigorous mathematical framework, setting it apart from many nature-inspired search algorithms. This paper proposes a hybrid optimization strategy based on CS-SDS techniques. By using the global search strategy solution of the SDS algorithm, this hybridization idea aims to enhance the cuckoo bird's search strategy for the optimum host nest. To that end, the SDS method would be used to place the cuckoo egg in the most advantageous location. When compared to other classifiers, PC2's improved performance may be attributed to its higher recall values. When compared to the Naive Bayes and Radial Bias Neural Network classifiers, the KNN performs 7.64% and 2.20% better, respectively.

Anomaly Detection in Data Center IT & Physical Infrastructure

Anomaly detection in data center IT and physical infrastructure is challenging due to the amount of heterogeneous data to be analyzed. Defining a solution that early identifies unexpected anomalies is particularly important to prevent data losses, breakdown of the system, and any other event considered to be critical for the activity of the data center. In the context of the INFN CNAF data center, one of the WLCG Tier-1s, we have performed a study based on monitored cooling, electrical, and IT hardware and software metrics to identify anomalies. In the present work, we aim to explore statistical approaches and machine learning solutions in the anomaly detection field for time series numerical metrics related to IT and physical infrastructure sensors. With the usage of statistical Z-score and percentile approaches and clustering DBSCAN technique, we have been able to group and identify anomalous events. Using the presented approach, different relevance can be attributed to the likely anomalies.

Software Quality Assessment Approach using Analytical Hierarchical Model: Applied in SMEs

Continuous software quality evolution is crucial albeit challenging due to tight budgets and timelines. In fact, neglecting software quality in favour of customer satisfaction; either by reducing internal or external defects or by adding technical features shall place much pressure on project managers and stakeholders to find a balance between project budget and schedule. The increasing demand for high-quality software among stakeholders emphasizes the need for comprehensive approaches that can evaluate and rank quality metrics to maximize their benefits. In this research paper, we introduce an assessment approach developed using key quality metrics integrated from Cost Of Software Quality (COSQ) metrics, reliability metrics, and cost of defects metrics based on quantitative multicriteria decision analysis capabilities of the Analytic Hierarchy Process (AHP) model applied with a real case study for the software development team in Small and Medium Enterprises (SMEs) since most of them do not follow standards nor adopt well-structured quality methods. The evaluation is established based on the integrated AHP assessment approach as a practical solution for organizations seeking a critical examination of the relative significance of software metrics in terms of their utility in enhancing business performance. As a result, an evaluation of these quality metrics has been conducted to highlight the percentage of weight for each quality metric based on interviews with subject matters experts. The ranking of software quality metrics greatly helps stakeholders in selecting the most appropriate attribute for evaluating the developed software to identify high-impact quality initiatives and measure their effectiveness in the Software Development Life Cycle (SDLC)

Improving Intelligence Metrics using Frequency Domain Convolutions for Improving Bug Prediction

Background: The novelty of the work lies in the formulation of these frequency-based generators, which reflects the lowest level of information loss in the intermediate calculations. The core idea behind the approach presented in this work is that a module with complex logic involved may have more probability of bugs. Software defect prediction is the area of research that enables the development and operations team to have the probability of bug proneness of the software. Many researchers have deployed multiple variations of machine learning and deep learning algorithms to achieve better accuracy and more insights into predictions. Objective: To prevent this fractional data loss from different derived metrics generations, a few optimal transformational engines capable of carrying forward formulations based on lossless computations have been deployed. Methods: A model Sodprhym has been developed to model refined metrics. Then, using some classical machine learning algorithms, accuracy measures have been observed and compared with the recently published results, which used the same datasets and prediction techniques. Results: The teams could establish watchdogs thanks to the automated detection, but it also gave them time to reflect on any potentially troublesome modules. For quality assurance teams, it has therefore become a crucial step. Software defect prediction looks forward to evaluating error-prone modules likely to contain bugs. Conclusion: Prior information can definitely align the teams with deploying more and more quality assurance checks on predicted modules. Software metrics are the most important component for defect prediction if we consider the different underlying aspects that define the defective module. Later we deployed our refined approach in which we targeted the metrics to be considered.

A catalog of metrics at source code level for vulnerability prediction: A systematic mapping study

AbstractIndustry practitioners assess software from a security perspective to reduce the risks of deploying vulnerable software. Besides following security best practice guidelines during the software development life cycle, predicting vulnerability before roll‐out is crucial. Software metrics are popular inputs for vulnerability prediction models. The objective of this study is to provide a comprehensive review of the source code‐level security metrics presented in the literature. Our systematic mapping study started with 1451 studies obtained by searching the four digital libraries from ACM, IEEE, ScienceDirect, and Springer. After applying our inclusion/exclusion criteria as well as the snowballing technique, we narrowed down 28 studies for an in‐depth study to answer four research questions pertaining to our goal. We extracted a total of 685 code‐level metrics. For each study, we identified the empirical methods, quality measures, types of vulnerabilities of the prediction models, and shortcomings of the work. We found that standard machine learning models, such as decision trees, regressions, and random forests, are most frequently used for vulnerability prediction. The most common quality measures are precision, recall, accuracy, and ‐measure. Based on our findings, we conclude that the list of software metrics for measuring code‐level security is not universal or generic yet. Nonetheless, the results of our study can be used as a starting point for future studies aiming at improving existing security prediction models and a catalog of metrics for vulnerability prediction for software practitioners.

State-of-the-art mathematically induced filtration approaches for smart spectrophotometric assessment of silodosin and solifenacin mixture in their new challenging formulation: Multi-tool greenness and whiteness evaluation

Benign prostatic hyperplasia is one of the most predominant health disorders in men with increasing incidence by age and usually accompanied with other bothersome symptoms. A new fixed dose combination, containing Silodosin and Solifenacin, has been recently launched for relieving such disorder associated with overactive bladder syndrome. In the current work, three smart, innovative and white spectrophotometric methods have been newly developed and optimized for simultaneous determination of the studied drugs in their binary mixture using water as an eco-friendly solvent. The adopted strategy relied on calculation of one or two factors as numerical constant or spectrum allowing mathematical filtration of desired analyte and full removal of any overlapped components in the mixture. The developed methods are categorized over two spectrophotometric platform windows. Window I deals with absorption spectra in its native forms (zero-order) including a newly developed method termed induced concentration subtraction (ICS) as well as induced dual wavelength (IDW) methods. Whereas window III is concerned with ratio spectra as in induced amplitude modulation (IAM) method. Compared to classical spectrophotometric methods, the proposed ones are superior in overcoming the inherited challenges in zero-order absorption spectrum of Solifenacin, particularly its very low absorptivity and lack of unique absorption maximum. Validity of the methods were thoroughly assured as per ICH guidelines with unified regression over 3.0–50.0 µg/mL in ICS method while IDW and IAM ones possessed linearity ranges of 3.0–50.0 µg/mL of Silodosin and 5.0–60.0 µg/mL of Solifenacin. The work was also extended to verify content uniformity of dosage units in accordance with USP recommendations. Greenness profile of the proposed methods was clearly assessed, in comparison to the reported analysis ones, via state-of-the-art software metrics, namely, green solvent selection tool (GSST), complementary green analytical procedure index (ComplexGAPI) and analytical greenness (AGREE). Finally, the proposed methods were in good adherence to the recently published postulates of white analytical chemistry.

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.

Software Reliability Prediction using Correlation Constrained Multi-Objective Evolutionary Optimization Algorithm

Software reliability frameworks are extremely effective for estimating the probability of software failure over time. Numerous approaches for predicting software dependability were presented, but neither of those has shown to be effective. Predicting the number of software faults throughout the research and testing phases is a serious problem. As there are several software metrics such as object-oriented design metrics, public and private attributes, methods, previous bug metrics, and software change metrics. Many researchers have identified and performed predictions of software reliability on these metrics. But none of them contributed to identifying relations among these metrics and exploring the most optimal metrics. Therefore, this paper proposed a correlation- constrained multi-objective evolutionary optimization algorithm (CCMOEO) for software reliability prediction. CCMOEO is an effective optimization approach for estimating the variables of popular growth models which consists of reliability. To obtain the highest classification effectiveness, the suggested CCMOEO approach overcomes modeling uncertainties by integrating various metrics with multiple objective functions. The hypothesized models were formulated using evaluation results on five distinct datasets in this research. The prediction was evaluated on seven different machine learning algorithms i.e., linear support vector machine (LSVM), radial support vector machine (RSVM), decision tree, random forest, gradient boosting, k-nearest neighbor, and linear regression. The result analysis shows that random forest achieved better performance.

Quality Analysis of Software Applications using Software Reliability Growth Models and Deep Learning Models

Finding the faults in the software is a very tedious task. Many software companies are trying to develop high-quality software which is having no faults. It is very important to analyze the errors, faults, and bugs in software development. Software reliability growth models (SRGM's) are used to help the software industries to create quality software products. Quality is the software metric that is used to analyze the performance of the software product. The software product which is having no errors or faults is considered the best software product. SRGM is also utilized to analyze the software quality based on the programming language. Deep Learning (DL) is a sub-domain in machine learning to solve several complex issues in software development. Finding accurate patterns from software faults is a very tedious task. DL algorithm performs better in integrating the SRGM with the DL approaches giving better results based on software fault detection. Many software faults real-time datasets are available to analyze the DL approaches. The performances of the various integrated models are analyzed by showing the quality metrics.

Implementation of artificial intelligence for the detection of cutaneous melanoma within a primary care setting: prevalence and types of skin cancer in outdoor enthusiasts.

There is enthusiasm for implementing artificial intelligence (AI) to assist clinicians detect skin cancer. Performance metrics of AI from dermoscopic images have been promising, with studies documenting sensitivity and specificity values equal to or superior to specialists for the detection of malignant melanomas (MM). Early detection rates would particularly benefit Australia, which has the worlds highest incidence of MM per capita. The detection of skin cancer may be delayed due to late screening or the inherent difficulty in diagnosing early skin cancers which often have a paucity of clinical features and may blend into sun damaged skin. Individuals who participate in outdoor sports and recreation experience high levels of intermittent ultraviolet radiation (UVR), which is associated with the development of skin cancer, including MM. This research aimed to assess the prevalence of skin cancer in individuals who regularly participate in activities outdoors and to report the performance parameters of a commercially available AI-powered software to assess the predictive risk of MM development. Cross-sectional study design incorporating a survey, total body skin cancer screening and AI-embedded software capable of predictive scoring of queried MM. A total of 423 participants consisting of surfers (n = 108), swimmers (n = 60) and walkers/runners (n = 255) participated. Point prevalence for MM was highest for surfers (6.48%), followed by walkers/runners (4.3%) and swimmers (3.33%) respectively. When compared to the general Australian population, surfers had the highest odds ratio (OR) for MM (OR 119.8), followed by walkers/runners (OR 79.74), and swimmers (OR 61.61) rounded out the populations. Surfers and swimmers reported comparatively lower lifetime hours of sun exposure (5,594 and 5,686, respectively) but more significant amounts of activity within peak ultraviolet index compared with walkers/runners (9,554 h). A total of 48 suspicious pigmented lesions made up of histopathology-confirmed MM (n = 15) and benign lesions (n = 33) were identified. The performance of the AI from this clinical population was found to have a sensitivity of 53.33%, specificity of 54.44% and accuracy of 54.17%. Rates of both keratinocyte carcinomas and MM were notably higher in aquatic and land-based enthusiasts compared to the general Australian population. These findings further highlight the clinical importance of sun-safe protection measures and regular skin screening in individuals who spend significant time outdoors. The use of AI in the early identification of MM is promising. However, the lower-than-expected performance metrics of the AI software used in this study indicated reservations should be held before recommending this particular version of this AI software as a reliable adjunct for clinicians in skin imaging diagnostics in patients with potentially sun damaged skin.

Using software metrics for predicting vulnerable classes in java and python based systems

ABSTRACT [Context:] Failure to predict vulnerability in the earlier stage of development can cause vulnerable code being written and deployed in the final software product. Vulnerability prediction using software metrics as features can support the discovery process by localizing vulnerable code. Existing studies have successfully employed metrics for vulnerability prediction for some platforms (C/C++ or Java projects). We propose that a comparative evaluation of how these metrics perform in projects of different languages can help the developers in deciding whether metrics-based prediction approach can be effective in their own project’s context. [Objective:] The purpose of this research is to analyze/compare the performance of software metrics in vulnerability-prediction for different programming language contexts (Java vs. Python). [Method:] We conducted experiments on vulnerabilities reported for Apache Tomcat (releases 6 and 7), Apache CXF, and two Python projects (Django and Keystone). We applied machine learning for predicting a particular type of code component (Java and Python classes) as vulnerable/non-vulnerable. [Results:] We found that metrics-based prediction can predict Java vulnerable classes with higher recall and precision than the Python vulnerable classes. [Conclusion:] This study at class-level will help developers to predict vulnerabilities at the class-level and assist in secure coding in object-oriented programming.

Software defects prediction by metaheuristics tuned extreme gradient boosting and analysis based on Shapley Additive Explanations

Software testing represents a crucial component of software development, and it is usually making the difference between successful and failed projects. Although it is extremely important, due to the fast pace and short deadlines of contemporary projects it is often neglected or not detailed enough due to the lack of available time, leading to the potential loss of reputation, private users’ data, money, and even lives in some circumstances. In such situations, it would be vital to have the option to predict what modules are error-prone according to the collection of software metrics, and to focus testing on them, and that task is a typical classification task. Machine learning models have been frequently employed within a wide range of classification problems with significant success, and this paper proposes eXtreme gradient boosting (XGBoost) model to execute the defect prediction task. A modified variant of the well-known reptile search optimization algorithm has been suggested to carry out the calibrating of the XGBoost hyperparameters. The enhanced algorithm was named HARSA and evaluated on the collection of challenging CEC2019 benchmark functions, where it exhibited excellent performance. Later, the introduced XGBoost model that uses the proposed algorithm has been evaluated on two benchmark software testing datasets, and the simulation outcomes have been compared to other powerful swarm intelligence metaheuristics that were used in the identical experimental environment, where the proposed approach attained superior classification accuracy on both datasets. Finally, Shapley Additive Explanations analysis was conducted to discover the impact of various software metrics on the classification results.

IMPLEMENTATION AND COMPARISON IN USING STATE PATTERN ON MAIN CHARACTER MOVEMENT (CASE STUDY : POCONG JUMP VIDEO GAME VERSION 1.0)

Game development success is often hard to achieve due to various problems such as performance issues, malfunctioning features, and poorly organized program structure. The problems that arise can be prevented by using the design pattern as a game programming architecture from the beginning of development. By implementing a design pattern, the process of developing video games can be made easier and simplified. The development team can focus its efforts on producing better quality video games. In this study, design patterns that would be used are state pattern and finite state machine. The state pattern is implemented by encapsulating the character's behavior in a class called state. Finite state machine will then facilitate the transition of states caused by user/player input or variable value changes. State pattern and finite state machine is tested with test case and game performance is tested with software metric. The result obtained from this study are state pattern and finite state machine have a valid component structure and could improve performance efficiency in video games.

Gaussian kernelized feature selection and improved multilayer perceptive deep learning classifier for software fault prediction

Software fault prediction is the significant process of identifying the errors or defects or faults in a software product. But, accurate and timely detection is the major challenging issue in different existing approaches to predicting software defects. A novel Gaussian linear feature embedding-based statistical test piecewise multilayer perceptive deep learning classifier (GLFE-STPMPDLC) is introduced to improve software fault prediction accuracy and minimize time consumption. First, the input data are collected from the dataset. Next, the software metrics selection is carried out to select the significant metrics using Gaussian kernelized locally linear embedding with lesser software fault prediction. Then classification is carried out by Kaiser Meyer piecewise multilayer perceptive deep learning classifier for software fault prediction. The novelty of Kaiser–Meyer–Olkin (KMO) correlation test analyzes testing and training instances. The innovation of the Heaviside step activation function is applied for analyzing the KMO correlation test results and providing the final software fault prediction results. Finally, accurate fault prediction outcomes are achieved at the output layer with lesser error. Simulation of proposed GLFE-STPMPDLC technique achieves better 5%, 3%, 3% and 3% enhancement of fault prediction accuracy, precision, recall, and f-measure and 13% faster prediction time compared to conventional methods.

Concept Drift in Software Defect Prediction: A Method for Detecting and Handling the Drift

Software Defect Prediction (SDP) is crucial towards software quality assurance in software engineering. SDP analyzes the software metrics data for timely prediction of defect prone software modules. Prediction process is automated by constructing defect prediction classification models using machine learning techniques. These models are trained using metrics data from historical projects of similar types. Based on the learned experience, models are used to predict defect prone modules in currently tested software. These models perform well if the concept is stationary in a dynamic software development environment. But their performance degrades unexpectedly in the presence of change in concept (Concept Drift). Therefore, concept drift (CD) detection is an important activity for improving the overall accuracy of the prediction model. Previous studies on SDP have shown that CD may occur in software defect data and the used defect prediction model may require to be updated to deal with CD. This phenomenon of handling the CD is known as CD adaptation. It is observed that still efforts need to be done in this direction in the SDP domain. In this article, we have proposed a pair of paired learners (PoPL) approach for handling CD in SDP. We combined the drift detection capabilities of two independent paired learners and used the paired learner (PL) with the best performance in recent time for next prediction. We experimented on various publicly available software defect datasets garnered from public data repositories. Experimentation results showed that our proposed approach performed better than the existing similar works and the base PL model based on various performance measures.

A novel approach for software defect prediction using CNN and GRU based on SMOTE Tomek method

Software defect prediction (SDP) plays a vital role in enhancing the quality of software projects and reducing maintenance-based risks through the ability to detect defective software components. SDP refers to using historical defect data to construct a relationship between software metrics and defects via diverse methodologies. Several prediction models, such as machine learning (ML) and deep learning (DL), have been developed and adopted to recognize software module defects, and many methodologies and frameworks have been presented. Class imbalance is one of the most challenging problems these models face in binary classification. However, When the distribution of classes is imbalanced, the accuracy may be high, but the models cannot recognize data instances in the minority class, leading to weak classifications. So far, little research has been done in the previous studies that address the problem of class imbalance in SDP. In this study, the data sampling method is introduced to address the class imbalance problem and improve the performance of ML models in SDP. The proposed approach is based on a convolutional neural network (CNN) and gated recurrent unit (GRU) combined with a synthetic minority oversampling technique plus the Tomek link (SMOTE Tomek) to predict software defects. To establish the efficiency of the proposed models, 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 experimental results showed that the proposed models predict the software defects more effectively on the balanced datasets than the original datasets, with an improvement of up to 19% for the CNN model and 24% for the GRU model in terms of AUC. We compared our proposed approach with existing SDP approaches based on several standard performance measures. The comparison results demonstrated that the proposed approach significantly outperforms existing state-of-the-art SDP approaches on most datasets.