Software Maintainability: Systematic Literature Review and Current Trends

Predicting the outcomes of software projects using data available in early development phases allows developers and managers more time to re-plan or re-design systems as needed to increase success. This research investigates early lifecycle predictive models of software faults and effort using information available no later than the system design phase. Static analysis of system designs characterizes intra- and inter-component control flow structure to calculate objective system design metrics. The model building process composes these design metrics into a set of relative factors that calibrate the relationships between the metrics and software faults and effort. These component-based predictive models provide a powerful approach to capture and analyze the designs of complex systems. These models facilitate the identification of essential system characteristics in terms of component structures and interactions, and they enable comprehensive analysis by being scalable to large systems. The research approach applies static analysis tools to the designs of 23 large-scale NASA systems containing over 5000 software components in order to automatically generate state- and interaction-based models of their designs. The static analysis tools derive the number of internal states for a component from its internal control flow design structure using cyclomatic complexity, and they calculate the number of interactions between components from the passing of control flow or data abstractions between components. The research approach evaluates the models by characterizing their relationships to favorable or unfavorable aspects of the system designs, in terms of component development faults and effort. The data analysis focuses on describing the characteristics of the state- and interaction-based models and identifying system design factors revealed by the models that relate to low or high faults, fault correction effort, and development effort. The results include the following, with comparative data being statistically significant at α < 0.05: ♦ The static analysis tools successfully generated state- and interaction-based models of the 23 system designs comprised of 5469 components. The numbers of states per component were exponentially distributed with an average of 14.3 and a median of 7.0. The numbers of interactions per component were exponentially distributed with an average of 8.8 and a median of 4.0. During system development, the components had 0.60 faults on average. ♦ Components with the most states (top 20%) had 7.8 times more faults than did those with the fewest states (bottom 20%). Components with the most states also had 22.5 times more fault correction effort and 5.4 times more development effort than did those with the fewest states. ♦ Components with the most interactions (top 20%) had 2.9 times more faults than did those with the fewest interactions (bottom 20%). Components with the most interactions also had 3.9 times more fault correction effort and 2.7 times more development effort than did those with the fewest interactions. In conclusion, we outline future research directions that build on these system modeling ideas and strategies.

Pharmacometrics is a multidisciplinary field utilizing mathematical models of physiology, pharmacology, and disease to describe and quantify the interactions between medication and patient. As these models become more and more advanced, the need for advanced data analysis tools grows. Recently, there has been much interest in the adoption of machine learning (ML) algorithms. These algorithms offer strong function approximation capabilities and might reduce the time spent on model development. However, ML tools are not yet an integral part of the pharmacometrics workflow. The goal of this work is to discuss how ML algorithms have been applied in four stages of the pharmacometrics pipeline: data preparation, hypothesis generation, predictive modelling, and model validation. We will also discuss considerations before the use of ML algorithms with respect to each topic. We conclude by summarizing applications that hold potential for adoption by pharmacometricians.

The study examines the prospects and challenges of machine learning (ML) applications in academic forecasting. Predicting academic activities through machine learning algorithms presents an enhanced means to accurately forecast academic events, including the academic performances and the learning style of students. The use of machine learning algorithms such as K-nearest neighbor (KNN), random forest, bagging, artificial neural network (ANN), and Bayesian neural network (BNN) has potentials that are currently being applied in the education sector to predict future events. Many gaps in the traditional forecasting techniques have greatly been bridged by the use of artificial intelligence-based machine learning algorithms thereby aiding timely decision-making by education stakeholders. ML algorithms are deployed by educational institutions to predict students' learning behaviours and academic achievements, thereby giving them the opportunity to detect at-risk students early and then develop strategies to help them overcome their weaknesses. However, despite the benefits associated with the ML approach, there exist some limitations that could affect its correctness or deployment in forecasting academic events, e.g., proneness to errors, data acquisition, and time-consuming issues. Nonetheless, we suggest that machine learning remains one of the promising forecasting technologies with the power to enhance effective academic forecasting that would assist the education industry in planning and making better decisions to enrich the quality of education.

To develop a prediction model that could risk stratify abdominal aortic aneurysms (AAAs) into high and low growth rate groups, using machine learning algorithms based on variables from different pathophysiological fields. A cohort of 40patients with small AAAs (maximum diameter 32-53mm) who had at least an initial and a follow-up CT scan (median follow-up 12months, range 3-36months) were included. 29 input variables from clinical, biological, morphometric, and biomechanical pathophysiological aspects extracted for predictive modeling. Collected data were used to build two supervised machine learning models. A gradient boosting (XGboost) and a support vector machines (SVM) algorithm were trained with 60% and tested with 40% of the data to predict which AAA would achieve a growth rate higher than the median of our study cohort. Receiver operating characteristics (ROC) curves and areas under the curve (AUC) were used for the evaluation of the developed algorithms. XGboost achieved the highest AUC in predicting high compared to low AAA growth rate with an AUC of 81.2% (95% CI from 61.1 to 100%). SVM achieved the second highest performance with an AUC of 68.8% (95% CI from 46.5 to 91%). Based on the best performing algorithm, variable importance was estimated. Diameter-diameter ratio (maximum diameter/neck diameter), Tortuosity from Renal arteries to aortic bifurcation, and maximum thickness of the intraluminal thrombus were found to be the most important factors for model predictions. Other factors were also found to play a significant but less important role. A prediction model that can risk stratify AAAs into high and low growth rate groups could be developed by analyzing several factors implicated in the multifactorial pathophysiology of this disease, with the use of machine learning algorithms. Future studies including larger patient cohorts and implementing additional risk markers may aid in the establishment of such methodology during AAA rupture risk estimation.

User interfaces are assuming an increasing importance within the design of a software product. This paper examines the various levels and components of window user interfaces and exposes their impact on software maintenance. It covers window systems, user interface toolkits, user interface management systems (UIMSs), dialogue separation and other user interface development concepts that will lead to more maintainable software. It also examines the research work on the testing of user interface software that is being pursued at the University of Durham.

This paper empirically investigates the relationships between 15 design metrics and maintainability of 148 Java open source software. The results show that size and complexity metrics are strongly related to the maintainability of open source software. However, cohesion and coupling, as currently captured by existing metrics, do not seem to have a significant impact on maintainability. When used together, these metrics can predict system maintainability fairly accurately (mean MREs below 30%).

Industrial chemical processes may involve thermal risks as most of the reactions performed are exothermic, the chemicals used are often thermally unstable, and the operating conditions are set to induce high conversion and throughput. Besides the reactive steps, all operations from mixing to storage and from processing to transport involving sensitive chemicals should be conducted under strictly controlled conditions ensuring safe operations. Performing an efficient risk assessment and implementing the proper risk mitigation measures are essential to avoid, or at least reduce, accidents and their potentially disastrous consequences. For optimal design and implementation of safety measures, it is important that these considerations are taken into account at early stages of process development. The required data should be made available at the appropriate time so it can be properly accounted for and efficiently serve the design. Yet, at early design phases, some information may be unavailable due to several reasons: the process design being still in development, some parameters can be unknown; experimental analysis of chemicals could be hindered or impossible due to products availability in required quantities; several alternatives are under investigation which raises the necessary resources (time and material) for experimental tests. Predictions would be the appropriate response to such scenario. The aim of this dissertation is to develop predictive models for two hazardous behaviors of chemicals: explosive sensitivity and thermal stability. For the models to be applicable at early development phases, it is preferable to minimize the information feed requirements, and therefore, structure-based approaches are applied. Two methods were identified: Quantitative Structure-Property Relationships (QSPR) and Group Contributions Method (GCM). The hazardous behaviors are studied through characteristic measurements: the Minimal Ignition Energy (MIE) to represent explosive sensitivity and Differential Scanning Calorimetry (DSC) for thermal stability. These measurements are widely employed in safety studies and deliver necessary information to identify potential hazards. Moreover, their specificities call for predictive models: MIE tests require repetitive analysis and hence are time and material consuming; regarding DSC experiments, experts have noted that they seem to exhibit structurally dependent features, and so far no study has comprehensively investigated this phenomenon. This work presents in a first part the structure-based approaches that are applied and the elements of Data Analysis necessary for developing predictive models and simulating experimental results. Secondly, both experimental analysis are detailed and the important information our models should be able to represent will be exposed. Finally, the third and last part is dedicated to the applications: the obtained predictive models are presented, evaluated and discussed. Most of the initial objectives are met as efficient solutions are proposed, nonetheless, some improvement strategies may also be considered.

The detection and identification of Distributed Denial-of-Service (DDoS) attacks remains a challenge in cloud/edge/fog computing environments. It usually requires network middleboxes, such as deep packet inspectors (DPI), for detection task mostly. But clouds and fogs have native powerful telemetry systems that are not yet fully exploited for DDoS detection; and provide so much information that could aid attack identification tasks as well. Machine Learning (ML) algorithms can help one diving into the richness of cloud’s native data collection services, which have a multitude of metrics from both physical and virtual hosts. This paper evaluates the use of ML algorithms over datasets collected from a experimental testbed based on OpenStack. Controlled attack scenarios were used to investigate the ability of ML for tasks such as detecting and identifying SYN_Flood and GET_Flood DDoS attacks mixed, in different proportions, with legitimate clients. kNN and Random Forest ML algorithms were trained and tested, and for evaluation the metrics accuracy, recall, precision, and F1-score were used. Our experiments presented about 87% of accuracy in the detection of SYN_Flood and GET_Flood DDoS attacks, whereas Snort IDS mostly fails to detect the latter attack by processing the corresponding packet traces. Also, the detection of PING_Flood DDoS attack was tested without training as an initial evaluation towards the generalization of the proposal.

In recent years, the use of machine learning algorithms to predict market trends has become a hot topic for researchers and enterprises. At present, time series prediction models, machine learning algorithms or deep learning algorithms are usually used to model and forecast the market of such cryptocurrencies. This paper proposes a cryptocurrency market prediction method based on CatBoost model and Bigdata. Firstly, the given data is preprocessed, and the dimension reduction of data features is mainly based on correlation analysis. Then build a basic model to fit the real data. Use test data to test and validate the generated base model, and adjust and optimize the model in the process. Finally, according to the given data, the optimized model is used for prediction. Experimental results show that the prediction model based on Catboost is superior to gradient propulsion, support vector machine and linear regression algorithm. Among them, metric value of Catboost model is 12% higher than Gradient Boosting model, 16.1% higher than SVM model, and 43.3% higher than linear regression model.

Maintainability is an essential dimension of software quality. Software Maintainability Prediction (SMP) is gaining the attention of researchers to develop maintainable software systems. Early prediction of software maintainability aid the software practitioners to focus on those software modules or classes that requires high maintainability effort in the maintenance phase. However, the imbalanced distribution of training data is a challenging and serious problem that is encountered while developing prediction models for software maintainability. This paper apply oversampling methods namely: Adaptive Synthetic Oversampling technique (AdaS), BorderlineSynthetic Minority Oversampling technique (BSMOTE), Synthetic Minority Oversampling technique (SMOTE), and SafeLevel Synthetic Minority Oversampling technique (SSMOTE) to treat the imbalanced data before learning the models for software maintainability. We also investigate the effectiveness of hybridized techniques for learning the prediction models using three popular Apache datasets. The outcome of the study supports the use of investigated oversampling methods with hybridized techniques to develop effective prediction models for software maintainability.

Background: This study proposes a cardiovascular diseases (CVD) prediction model using machine learning (ML) algorithms based on the National Health Insurance Service-Health Screening datasets. Methods: We extracted 4699 patients aged over 45 as the CVD group, diagnosed according to the international classification of diseases system (I20–I25). In addition, 4699 random subjects without CVD diagnosis were enrolled as a non-CVD group. Both groups were matched by age and gender. Various ML algorithms were applied to perform CVD prediction; then, the performances of all the prediction models were compared. Results: The extreme gradient boosting, gradient boosting, and random forest algorithms exhibited the best average prediction accuracy (area under receiver operating characteristic curve (AUROC): 0.812, 0.812, and 0.811, respectively) among all algorithms validated in this study. Based on AUROC, the ML algorithms improved the CVD prediction performance, compared to previously proposed prediction models. Preexisting CVD history was the most important factor contributing to the accuracy of the prediction model, followed by total cholesterol, low-density lipoprotein cholesterol, waist-height ratio, and body mass index. Conclusions: Our results indicate that the proposed health screening dataset-based CVD prediction model using ML algorithms is readily applicable, produces validated results and outperforms the previous CVD prediction models.

The widespread implementation of smart meters (SM) and the deployment of the advanced metering infrastructure (AMI) provide large amounts of fine-grained data on prosumers. Machine learning (ML) algorithms are used in different techniques, e.g. non-intrusive load monitoring (NILM), to extract useful information from collected data. However, the use of ML algorithms to gain insight on prosumer behavior and characteristics raises not only numerous technical but also legal concerns. This paper maps electricity prosumer concerns towards the AMI and its ML based analytical tools in terms of data protection, privacy and cybersecurity and conducts a legal analysis of the identified prosumer concerns within the context of the EU regulatory frameworks. By mapping the concerns referred to in the technical literature, the main aim of the paper is to provide a legal perspective on those concerns. The output of this paper is a visual tool in form of a table, meant to guide prosumers, utility, technology and energy service providers. It shows the areas that need increased attention when dealing with specific prosumer concerns as identified in the technical literature.

28 - Structural monitoring of adhesive joints using machine learning

&amp;lt;p&amp;gt;The machine learning (ML) algorithms have already found their application in standard seismological procedures, such as earthquake detection and localization, phase picking, earthquake early warning system, etc. They are progressively becoming superior methods since one can rapidly scan voluminous data and detect earthquakes, even if buried in highly noisy time series.&amp;lt;/p&amp;gt;&amp;lt;p&amp;gt;We here make use of ML algorithms to obtain more complete near fault seismic catalogs and thus better understand the long-term (decades) evolution of seismicity before large earthquakes occurrence. We focus on data recorded before the devastating L&amp;amp;#8217;Aquila earthquake (6 April 2009 01:32 UTC, Mw6.3) right beneath the city of L&amp;amp;#8217;Aquila in the Abruzzo region (Central Italy). Before this event sparse stations were available, reducing the magnitude completeness of standard catalogs.&amp;amp;#160;&amp;lt;/p&amp;gt;&amp;lt;p&amp;gt;We adapted existing convolutional neural networks (CNN) for earthquake detection, localisation and characterization using a single-station waveforms. The CNN is applied to 29 years of data (1990 to 2019) recorded at the AQU station, located near the city of L&amp;amp;#8217;Aquila (Italy). The pre-existing catalog maintained by Istituto nazionale di geofisica e vulcanologia is used to define labels and train and test the CNN. We are here interested in classifying the continuous three-component waveforms into four categories, noise/earthquakes, distance (location), magnitude and depth, where each category consists of several nodes. Existing seismic catalogs are used to label earthquakes, while the noise events are randomly selected between the catalog events, evenly represented by daytime and night-time periods.&amp;lt;/p&amp;gt;&amp;lt;p&amp;gt;We prefer CNN over other methods, since we can use seismograms directly with very minor pre-processing (e.g. filtering) and we do not need any prior knowledge of the region.&amp;lt;/p&amp;gt;&amp;lt;p&amp;gt;&amp;lt;br&amp;gt;&amp;lt;br&amp;gt;&amp;lt;/p&amp;gt;

Capitalization of lessons learned on Asset Integrity Management during Front End Loading phases of a green field Project Development, by defining plan for implementation of a diagnostic digital tool for reducing downtime and introduce predictive maintenance during Operation. Eni developed a platform of Digital applications for enhanced Operations management by implementing an Integrated Asset Management (IAM) system. Advanced Analytics tool is part of it and is designed for monitoring, foreseeing and preventing production upsets and anomalies; the tool is set up by verification of areas of interest and criticalities, with identification of main equipment data sets and by the implementation and validation of predictive models. Starting from historical data, data scientists supported by experts develop algorithms capable of finding interdependencies between a set of input variables and an output variable (phenomenon to be predicted/monitored), thus detecting anomalies and criticalities. Main areas of benefit are envisaged on Production continuity, capable of predicting problems on static and rotating equipment and giving information on the most impacting variables on the incipient problems. The tool will support technicians to help them preventing failures and out-of-specs events which may cause loss of production or asset integrity issues, with the activation of predictive maintenance and the aim to strive a continuous monitoring and improvement of plant operational performances. An Energy Efficiency predictive model will also be set up, capable of forecasting the future energy performances of the asset through the prediction of the Stationary Combustion of Carbon Dioxide (CO2) emission index (t CO2/kbbl) and providing the list of the main influencing equipment and variables. The plan for implementation of the tool from the Early phases of development help the organization on prioritizing the implementation of Digital tools as part of the execution and realization of the Asset to be delivered to the Operational personnel, by easing the transition and avoiding subsequent retrofitting carrying brownfield works and additional costs. The implementation of Advanced Analytics tool has been embedded in a new green field initiative of a Development Project since Front End Loading phases, thus fostering digital implementation and minimizing deployment costs by including those as part of the Investment Proposal presented to Joint Venture Partners and Authorities.