A. Güneş Koru

An investigation of the effect of module size on defect prediction using static measures

We used several machine learning algorithms to predict the defective modules in five NASA products, namely, CM1, JM1, KC1, KC2, and PC1. A set of static measures were employed as predictor variables. While doing so, we observed that a large portion of the modules were small, as measured by lines of code (LOC). When we experimented on the data subsets created by partitioning according to module size, we obtained higher prediction performance for the subsets that include larger modules. We also performed defect prediction using class-level data for KC1 rather than the method-level data. In this case, the use of class-level data resulted in improved prediction performance compared to using method-level data. These findings suggest that quality assurance activities can be guided even better if defect prediction is performed by using data that belong to larger modules.

Identifying and characterizing change-prone classes in two large-scale open-source products

Developing and maintaining open-source software has become an important source of profit for many companies. Change-prone classes in open-source products increase project costs by requiring developers to spend effort and time. Identifying and characterizing change-prone classes can enable developers to focus timely preventive actions, for example, peer-reviews and inspections, on the classes with similar characteristics in the future releases or products. In this study, we collected a set of static metrics and change data at class level from two open-source projects, KOffice and Mozilla. Using these data, we first tested and validated Paretos Law which implies that a great majority (around 80%) of change is rooted in a small proportion (around 20%) of classes. Then, we identified and characterized the change-prone classes in the two products by producing tree-based models. In addition, using tree-based models, we suggested a prioritization strategy to use project resources for focused preventive actions in an efficient manner. Our empirical results showed that this strategy was effective for prioritization purposes. This study should provide useful guidance to practitioners involved in development and maintenance of large-scale open-source products.

Theory of relative defect proneness

In this study, we investigated the functional form of the size-defect relationship for software modules through replicated studies conducted on ten open-source products. We consistently observed a power-law relationship where defect proneness increases at a slower rate compared to size. Therefore, smaller modules are proportionally more defect prone. We externally validated the application of our results for two commercial systems. Given limited and fixed resources for code inspections, there would be an impressive improvement in the cost-effectiveness, as much as 341% in one of the systems, if a smallest-first strategy were preferred over a largest-first one. The consistent results obtained in this study led us to state a theory of relative defect proneness (RDP): In large-scale software systems, smaller modules will be proportionally more defect-prone compared to larger ones. We suggest that practitioners consider our results and give higher priority to smaller modules in their focused quality assurance efforts.

An empirical characterization of scientific software development projects according to the Boehm and Turner model: A progress report

A number of recent studies reported on the success of applying agile methods in scientific software development projects. These studies found that agile methods are well suited to the exploratory, iterative, and collaborative nature of scientific inquiry. However, these findings might not be applicable in all situations pertaining to scientific software development projects. In addition, they only constitute a subset of the important factors while deciding which development methods and practices should be adopted. Therefore, it becomes important to conduct further research before making recommendations regarding the adoption of certain development methods and practices in this domain. In this progress report, we discuss our on-going research that will empirically study the characteristics of various scientific software development projects according to a model suggested by Boehm and Turner. We plan to conduct interviews and collect data from various scientific software development projects in the Baltimore-Washington area. We expect that our qualitative results will increase our understanding of the characteristics in those projects favoring plan-driven approaches or agile methods, and the needs and conditions associated with those characteristics. Our research provides guidance to scientific software developers by enhancing their capacity to evaluate and understand their own project characteristics and select effective software practices. As a long-term benefit in the same direction, our qualitative results will generate a set of hypotheses that can be tested in different project environments to better understand and categorize scientific software development projects. Consequently, in the future, more generalizable and actionable recommendations can be made for scientific software development projects.

Software Engineering Education for Bioinformatics

As software engineering educators, it is important for us to realize the increasing domain-specificity of software, and incorporate these changes in our design of teaching material. Bioinformatics software is an example of immensely complex and critical scientific software and this domain provides an excellent illustration of the role of computing in the life sciences. To study bioinformatics from a software engineering standpoint, we conducted an exploratory survey of bioinformatics developers. The survey had a range of questions about people, processes and products. We learned that practices like extreme programming, requirements engineering and documentation. As software engineering educators, we realized that the survey results had important implications for the education of bioinformatics professionals. We also investigated the current status of software engineering education in bioinformatics, by examining the curricula of more than fifty bioinformatics programs and the contents of over fifteen textbooks. We observed that there was no mention of the role and importance of software engineering practices essential for creating dependable software systems. Based on our findings and existing literature we present a set of recommendations for improving software engineering education in bioinformatics.

Alcohol-Related Diagnoses in Hospital Admissions for All Causes Among Middle-Aged and Older Adults: Trends and Cohort Differences From 1993 to 2010

Objective: This aim of this study was to characterize trends in alcohol-related hospital admissions among middle-aged and older adults from 1993 to 2010 in relation to age, gender, race, and cohort membership. Method: This study utilized repeated cross-sectional data from the Nationwide Inpatient Sample. Using alcohol-related classified admissions, yearly rates and longitudinal trends of alcohol-related inpatient hospitalizations based on age, period, birth cohort, gender, and race were estimated. Results: Among those aged 45 and older, admissions rose from an estimated 610,634 to more than 1,134,876, and rates of any alcohol-related diagnosis also increased from 1993 to 2010. Rates for men were consistently higher than women, and rates for Blacks were higher than Whites. Age was associated with decreasing rates, but post–World War II cohorts displayed higher rates over time. Discussion: Rates of alcohol-related admissions are increasing among adults above age 45, which may be a function of cohort effects. Training the health care workforce is crucial to respond to this trend.

Teaching software engineering to end-users

Bioinformatics software is an example of immensely complex and critical scientific software, and this domain provides an excellent illustration of the role of end user computing in the sciences. To explore these interesting characteristics from a software engineering standpoint, we had conducted an exploratory survey of bioinformatics developers. The survey had a range of questions about people, processes and products. As software engineering researchers, we realized that the survey results had important implications for the education of bioinformatics software professionals. Through this paper we intend to open an avenue of discussion about software engineering knowledge that should be taught to end user programmers, based on our findings in the bioinformatics domain. In addition to the survey results we went through the curricula of more than fifty bioinformatics programs as well as the contents of over fifteen textbooks. We observed that there was no mention of the role and importance of software engineering practices essential for creating dependable software systems. We present a set of recommendations for improving bioinformatics education in terms of software engineering principles and ways that they apply in the context of end-user development.

The effect of human memory organization on code reviews under different single and pair code reviewing scenarios

Human memory organization has been shown to be related to how programmers understand programs. In recent years, agile methods brought the emphasis back on human and social aspects of software engineering with a set of new principles and practices. One of them, pair programming has been shown to improve quality and reduce the development costs. In this position paper, we propose a controlled experiment to evaluate the effect of human memory organization through chunking on code reviews under different single and pair code reviewing scenarios.

Theory of Relative Dependency: Higher Coupling Concentration in Smaller Modules and its Implications for Software Refactoring and Quality

Our recent studies have repeatedly found that smaller modules are proportionally more defect-prone. In this paper, we formulated and tested a hypothesis stating that smaller modules are proportionally more coupled given that dependencies caused by coupling has been consistently associated with defect-proneness. We found strong evidence supporting this hypothesis. Further, we found that refactoring practices exacerbate this effect. Based on the highly consistent results obtained in this study, we state an empirically-based theory for software modules called the theory of relative dependency: In large scale software systems, smaller modules will be proportionally more dependent compared to larger ones. The implications of our findings for practice are twofold: (1) we now have an empirically supported mechanism explaining the observations that defect concentration is higher in smaller modules, which can be used by practitioners as an evidence while seeking resources and support to revise or amend the existing quality assurance and quality control practices in their organizations; (2) for projects that refactor extensively, such as those using agile methods, focusing defect detection and correction activities on smaller modules will lead to even more effective defect detection.

A tree-based approach to preserve the privacy of software engineering data and predictive models

In empirical disciplines, data sharing leads to verifiable research and facilitates future research studies. Recent efforts of the PROMISE community contributed to data sharing and reproducible research in software engineering. However, an important portion of data used in empirical software engineering research still remains classified. This situation is unlikely to change because many companies, governments, and defense organizations will be always hesitant to share their project data such as, effort and defect data, due to various confidentiality, privacy, and security concerns. In this paper, we present, demonstrate, and evaluate a novel tree-based data perturbation approach. This approach does not only preserve privacy effectively, but it also preserves the predictive patterns in the original data set. Consequently, the empirical software engineering researchers will have access to another category of data sets, transformed data sets, which will increase the verifiability of research results and facilitate the future research studies in this area. Our approach can be immediately useful to many researchers and organizations who are willing to share their software engineering data but cannot do so due to privacy concerns.