Professor Rachel Harrison

The research area of Software Defect Prediction (SDP) is both extensive and popular, and is often treated as a classification problem. Improvements in classification, pre-processing and tuning techniques, (together with many factors which can influence model performance) have encouraged this trend. However, no matter the effort in these areas, it seems that there is a ceiling in the performance of the classification models used in SDP. In this paper, the issue of classifier performance is analysed from the perspective of data complexity. Specifically, data complexity metrics are calculated using the Unified Bug Dataset, a collection of well-known SDP datasets, and then checked for correlation with the defect prediction performance of machine learning classifiers (in particular, the classifiers C5.0, Naive Bayes,  Artificial Neural Networks, Random Forests, and Support Vector Machines). In this work, different domains of competence and incompetence are identified for the classifiers. Similarities and differences between the classifiers and the performance metrics are found and the Unified Bug Dataset is analysed from the perspective of data complexity. We found that certain classifiers work best in certain situations and that all data complexity metrics can be problematic, although certain classifiers did excel in some situations.

Importance: Usability is an important method for evaluating mobile health apps from a user perspective. However, numerous apps are available to the public with insufficient effort devoted to their design, development, and evaluation.

Objective: To assess usability evaluation and reporting for mobile health apps targeting patients with skin diseases.

Evidence Review: The protocol was registered with PROSPERO (CRD42022347184). A search strategy combined terms for usability evaluation, user experience, skin disease and mobile health apps (search date from 2012 to May 27, 2022). Six databases (Embase, Medline, CINAHL, Cochrane CENTRAL, Cochrane Database of Systematic Reviews, Scopus) were searched identifying 15234 results. Nine studies (comprising 9 apps) were included in the final analysis.

Findings: Skin diseases targeted included sun protection (n=3), skin cancer (n=3), and one study identified for each of the following: cutaneous leishmaniasis, spina bifida, chronic pruritus, acne, psoriasis, rosacea laser treatments, actinic damage, monitoring benign moles, alopecia, and inflammatory rash. All studies assessed app usability and/or feasibility, with 8/9 studies concluding their app was useful and easy to use. Most studies were qualitative employing usability questionnaires or semi-structured interviews. Common themes emerging included: easy to use/easy to navigate, easy to understand, security/privacy and data sharing issues, customizability, costs and ability to track progress / self-monitor. The most common modules in the apps included taking/storing skin images on the app, skin check reminders, education modules and e-diaries/rating of symptoms. Usability encompassed various concepts including feasibility, usefulness, and satisfaction.

Conclusion and Relevance: Smartphone applications used in the context of skin disorders, are being developed across multiple diseases and indications and when evaluated appear to have a high degree of usability. However there are concerns given the paucity of reported literature compared with the current explosion of app development and deployment which highlights there is a need for careful user-centred design and rigorous evaluation of these tools before they are released for widespread uptake. Development of any public facing app especially those targeting skin diseases should aim to evaluate usability and report these findings  to optimise longer-term uptake of these apps.

Context. Software is subject to the presence of faults, which impacts its quality as well as production and maintenance costs. Evolutionary fault localization uses data from the test activity (test spectra) as a source of information about defects, and its automation aims to obtain better accuracy and lower software repair cost.

Motivation. Our analysis identified that the test spectra commonly used in research exhibit a high ratio of sample repetition, which impairs the training and evolution of Genetic Programming (GP) evolved heuristics.

Problem. We investigate whether evolutionary training based on the uniqueness of suspiciousness scores can increase the ability to find software faults (defects), even in repeat-sample scenarios. Specifically, we examine whether the GP-evolved heuristic, which is based on the distinguishability of program elements in terms of faults, is really competitive.

Methodology. The investigation formalized hypotheses, introduced three training strategies to guide the research and carried out an experimental evaluation, aiming to reach conclusions regarding the assessment of research questions and hypotheses.

Analysis. The results have shown the competitiveness of all the proposed training strategies through evaluation metrics commonly used in the research field.

Conclusion. Statistical analyses confirmed that the uniqueness of suspiciousness scores guides the generation of superior heuristics for fault localization.

Much has been written about the algorithmic role that AI plays for automation in SE.
But what about the role of AI, augmented by human knowledge? Can we make a profound
advance by combining human and artificial intelligence? Researchers in requirements
engineering think so, arguing that requirement engineering is the secret weapon for better AI
and better software.

Context. Software Fault Localisation (FL) refers to finding faulty software elements related to failures produced as a result of test case execution. This is a laborious and time consuming task. To allow FL automation search-based algorithms have been successfully applied in the field of Search-Based Fault Localisation (SBFL). However, there is no study mapping the SBFL field to the best of our knowledge and we believe that such a map is important to promote new advances in this field. Objective. To present the results of a mapping study on SBFL, by characterising the proposed methods, identifying sources of used information, adopted evaluation functions, applied algorithms and elements regarding reported experiments. Method. Our mapping followed a defined process and a search protocol. The conducted analysis considers different dimensions and categories related to the main characteristics of SBFL methods. Results. All methods are grounded on the coverage spectra category. Overall the methods search for solutions related to suspiciousness formulae to identify possible faulty code elements. Most studies use evolutionary algorithms, mainly Genetic Programming, by using a single-objective function. There is little investigation of real-and-multiple-fault scenarios, and the subjects are mostly written in C and Java. No consensus was observed on how to apply the evaluation metrics. Conclusions. Search-based fault localisation has seen a rise in interest in the past few years and the number of studies has been growing. We identified some research opportunities such as exploring new sources of fault data, exploring multi-objective algorithms, analysing benchmarks according to some classes of faults, as well as, the use of a unique definition for evaluation measures.

This paper presents an overview of the usability engineering process for the development of a personalised clinical decision support system for the management of type 1 diabetes. The tool uses artificial intelligence (AI) techniques to provide insulin bolus dose advice and carbohydrate recommendations that adapt to the individual. We describe the role of human factors and user-centred design in the creation of medical systems that must adhere to international standards. We focus specifically on the formative evaluation stage of this process. The preliminary analysis of data shows promising results.

Background: People with T1DM are confronted with self-management tasks and for which they need to develop strategies to balance the risks of long-term complications with those of hypoglycemic events. The potential of advanced and evolving technology to address these issues involves consideration of psychological and behavioral constructs alongside evaluation of the usability of devices. Access and uptake of advanced technology is further influenced by economic factors and health care provider capacity to support such interventions. Previous reviews have either focused upon clinical outcomes or descriptively scoped the literature. In addition, some have synthesized studies on adults with those on children and young people where human factors are different. Objective: The objective of this review was to describe the relationship between

human factors and adherence with technology for data logging processes in adults with T1DM and to explore the factors which influence this association. Methods: A systematic search of the literature was undertaken in accordance with the PRISMA guidelines. Quality appraisal of each study was undertaken. Data were abstracted and categorized into the themes that underpinned the human factor constructs that were examined. Results: Eighteen studies were included in the review. Six constructs emerged from the data analysis: The relationship between adherence to data logging and measurable outcomes; Satisfaction with the transition to advanced technology for self-management; Use of advanced technology and time spent on diabetes related activities; Strategies to mediate the complexities of diabetes and the use of advanced technology; Cognition in the wild and, Meanings, views and perspectives from the users of technology. Conclusions: evidence of increased treatment satisfaction was found on transition from traditional to advanced technology use (insulin pump and continuous glucose monitoring (CGM)); the most significant contributing factor was when blood glucose (BG) levels were consistently

evidence that logging of data was positively correlated with increasing age when using an app that provided meaningful feedback (regression coefficient = 55.8 recordings/ year; P = 0.009). Furthermore, there were benefits of CGM for older people in mediating complexities and their fears of hypoglycemia with reported significant differences in well-being (P= .009). Qualitative studies within the review aimed to explore the use and uptake of technology within the context of everyday lives. There were ‘frustrations’ with CGM, continuous subcutaneous insulin infusion (CSII), calibration of devices and alarms. This created implications for “body image” and the way in which “significant others” impacted on the behavior and attitude of the individual towards technology use. There were wide variations in the normal use of and interaction with technology across a continuum of sociocultural contexts, which has implications for the way in which future technologies should be designed. Many of the quantitative studies in the review were limited by small sample sizes. This may make it difficult to generalize findings to other contexts. This is further limited by a sample that was predominantly Caucasian, well-controlled and engaged with their self-care. However, the use of critical appraisal frameworks has highlighted areas where research into human factors and data logging processes of individuals could be improved. This includes engaging people in the design of the technology especially hard-to- reach or marginalized groups.

Context

Although many papers have been published on software defect prediction techniques, machine learning approaches have yet to be fully explored.

Objective

In this paper we suggest using a descriptive approach for defect prediction rather than the precise classification techniques that are usually adopted. This allows us to characterise defective modules with simple rules that can easily be applied by practitioners and deliver a practical (or engineering) approach rather than a highly accurate result.

Method

We describe two well-known subgroup discovery algorithms, the SD algorithm and the CN2-SD algorithm to obtain rules that identify defect prone modules. The empirical work is performed with publicly available datasets from the Promise repository and object-oriented metrics from an Eclipse repository related to defect prediction. Subgroup discovery algorithms mitigate against characteristics of datasets that hinder the applicability of classification algorithms and so remove the need for preprocessing techniques.

Results

The results show that the generated rules can be used to guide testing effort in order to improve the quality of software development projects. Such rules can indicate metrics, their threshold values and relationships between metrics of defective modules.

Conclusions

The induced rules are simple to use and easy to understand as they provide a description rather than a complete classification of the whole dataset. Thus this paper represents an engineering approach to defect prediction, i.e., an approach which is useful in practice, easily understandable and can be applied by practitioners.