Megha Khanna

Investigation of relationship between object-oriented metrics and change proneness

Software is the heartbeat of modern day technology. In order to keep up with the pace of modern day expansion, change in any software is inevitable. Defects and enhancements are the two main reasons for a software change. The aim of this paper is to study the relationship between object oriented metrics and change proneness. Software prediction models based on these results can help us identify change prone classes of a software which would lead to more rigorous testing and better results. In the previous research, the use of machine learning methods for predicting faulty classes was found. However till date no study determines the effectiveness of machine learning methods for predicting change prone classes. Statistical and machine learning methods are two different techniques for software quality prediction. We evaluate and compare the performance of these machine learning methods with statistical method (logistic regression). The results are based on three chosen open source software, written in java language. The performance of the predicted models was evaluated using receiver operating characteristic analysis. The study shows that machine learning methods are comparable to regression techniques. Testing based on change proneness of a software leads to better quality by targeting the most change prone classes. Thus, the developed models can be used to reduce the probability of defect occurrence and we commit ourselves to better maintenance.

On the application of search-based techniques for software engineering predictive modeling: A systematic review and future directions

Abstract Software engineering predictive modeling involves construction of models, with the help of software metrics, for estimating quality attributes. Recently, the use of search-based techniques have gained importance as they help the developers and project-managers in the identification of optimal solutions for developing effective prediction models. In this paper, we perform a systematic review of 78 primary studies from January 1992 to December 2015 which analyze the predictive capability of search-based techniques for ascertaining four predominant software quality attributes, i.e., effort, defect proneness, maintainability and change proneness . The review analyses the effective use and application of search-based techniques by evaluating appropriate specifications of fitness functions, parameter settings, validation methods, accounting for their stochastic natures and the evaluation of developmental models with the use of well-known statistical tests. Furthermore, we compare the effectiveness of different models, developed using the various search-based techniques amongst themselves, and also with the prevalent machine learning techniques used in literature. Although there are very few studies which use search-based techniques for predicting maintainability and change proneness, we found that the results of the application of search-based techniques for effort estimation and defect prediction are encouraging. Hence, this comprehensive study and the associated results will provide guidelines to practitioners and researchers and will enable them to make proper choices for applying the search-based techniques to their specific situations.

Mining the impact of object oriented metrics for change prediction using Machine Learning and Search-based techniques

Change in a software is crucial to incorporate defect correction and continuous evolution of requirements and technology. Thus, development of quality models to predict the change proneness attribute of a software is important to effectively utilize and plan the finite resources during maintenance and testing phase of a software. In the current scenario, a variety of techniques like the statistical techniques, the Machine Learning (ML) techniques and the Search-based techniques (SBT) are available to develop models to predict software quality attributes. In this work, we assess the performance of ten machine learning and search-based techniques using data collected from three open source software. We first develop a change prediction model using one data set and then we perform inter-project validation using two other data sets in order to obtain unbiased and generalized results. The results of the study indicate comparable performance of SBT with other employed statistical and ML techniques. This study also supports inter project validation as we successfully applied the model created using the training data of one project on other similar projects and yield good results.

Examining the effectiveness of machine learning algorithms for prediction of change prone classes

Managing change in the early stages of a software development life cycle is an effective strategy for developing a good quality software at low costs. In order to manage change, we use software quality models which can efficiently predict change prone classes and hence guide developers in appropriate distribution of limited resources. This study examines the effectiveness of ten machine learning algorithms for developing such software quality models on three object-oriented software data sets. We also compare the performance of machine learning algorithms with the widely used logistic regression technique and statistically rank various algwith the help of Friedman test.

Analyzing software change in open source projects using Artificial Immune System algorithms

Development of software change prediction models, based on the change histories of a software, are valuable for early identification of change prone classes. Classification of these change prone classes is vital to yield competent use of limited resources in an organization. This paper validates Artificial Immune System (AIS) algorithms for development of change prediction models using six open source data sets. It also compares the performance of AIS algorithms with other machine learning and statistical algorithms. The results of the study indicate, that the models developed, are effective means of predicting change prone classes in the future versions of the software. However, AIS algorithms do not perform better that machine learning and other statistical algorithms. The study provides conclusive results about the capabilities of AIS algorithms and reports whether there are any significant differences in the performance of different algorithms using a statistical test.

Applicability of Inter Project Validation for Determination of Change Prone Classes

research in the field of defect and change proneness prediction of software has gained a lot of momentum over the past few years. Indeed, effective prediction models can help software practitioners in detecting the change prone modules of a software, allowing them to optimize the resources used for software testing. However, the development of the prediction models used to determine change prone classes are dependent on the availability of historical data from the concerned software. This can pose a challenge in the development of effective change prediction models. The aim of this paper is to address this limitation by using the data from models based on similar projects to predict the change prone classes of the concerned software. This inter project technique can facilitate the development of generalized models which can be used to ascertain change prone classes for multiple software projects. It would also lead to optimization of critical time and resources in the testing and maintenance phases. This work evaluates the effectiveness of statistical and machine learning techniques for developing such models using receiver operating characteristic analysis. The observations of the study indicate varied results for the different techniques used.

Particle swarm optimization-based ensemble learning for software change prediction

Abstract Context Various researchers have successfully established the association between Object-Oriented metrics and change prone nature of a class. However, they actively continue to explore effective classifiers for developing efficient change prediction models. Recent developments have ascertained that ensemble methodology can be used to improve the prediction performance of individual classifiers. Objective This study proposes four strategies of ensemble learning to predict change prone classes by combining seven individual Particle Swarm Optimization (PSO) based classifiers as constituents of ensembles and aggregating them using weighted voting. Method The weights allocated to individual classifiers are based on their accuracy and their ability to correctly predict “hard instances” i.e. classes which are frequently misclassified by a majority of classifiers. Each individual PSO based classifier uses a different fitness function. The ensembles are constructed on the premises that change in fitness functions leads to variation in the results of a search-based algorithm such as PSO. Therefore, it is important to combine them to obtain a better classifier with improved accuracy using the ensemble methodology. Results The proposed strategies of ensemble learning were found effective in predicting software change. The statistical analysis of the results indicates improved performance of the proposed ensemble classifiers as compared to individual classifiers. Furthermore, the results of the proposed voting ensemble classifiers were found competent with those of machine-learning ensemble classifiers for determination of change prone classes. Conclusion The accuracy and diversity of the individual classifiers were instrumental in the superior performance of the proposed voting ensemble classifiers.

Threats to validity in search-based predictive modelling for software engineering

A number of studies in the literature have developed effective models to address prediction tasks related to a software product such as estimating its development effort, or its change/defect proneness. These predictions are critical as they help in identifying weak areas of a software product and thus guide software project managers in effective allocation of project resources to these weak parts. Such practices assure good quality software products. Recently, the use of search-based approaches (SBAs) for developing software prediction models (SPMs) has been successfully explored by a number of researchers. However, in order to develop effective and practical SPMs it is imperative to analyse various sources of threats. This study extensively reviews 93 primary studies, which use SBAs for developing SPMs of four commonly used software attributes (effort, defect-proneness, maintainability and change-proneness) in order to discuss and identify the various sources of threats while using these approaches for SPMs. The study also lists various actions that may be taken in order to minimise these threats. Furthermore, best practice examples in literature and the year-wise trends of threats indicating the most common threats missed by researchers are provided to help academicians and practitioners in designing effective studies for developing SPMs using SBAs.