Machine Learning Software Architecture and Model Workflow. A Case of Django REST Framework

[Context/Background] Machine Learning (ML) software has special ability for increasing technical debt due to ML-specific issues besides having all the problems of regular code. The term “Hidden Technical Debt” (HTD) was coined by Sculley et al. to address maintainability issues in ML software as an analogy to technical debt in traditional software. [Goal] The aim of this paper is to empirically analyse how HTD patterns emerge during the early development phase of ML software, namely the prototyping phase. [Method] Therefore, we conducted a case study with subject systems as ML models planned to be integrated into the software system owned by Vasttrafik, the public transportation agency in the west area of Sweden. [Results] During our case study, we could detect HTD patterns, which have the potential to emerge in ML prototypes, except for “Legacy Features”, “Correlated features”, and “Plain Old Data Type Smell”. [Conclusion] Preliminary results indicate that emergence of significant amount of HTD patterns can occur during prototyping phase. However, generalizability of our results require analyses of further ML systems from various domains.

This Special Collection calls for contributions that advance software as medical device applications for the screening, early detection, differential diagnosis, prognosis and improved clinical management of cancers. With the increasing growth in data in medicine, the development of computational methods and workflows to effectively harness this resource and turn it into actionable information is in great need. Artificial Intelligence (including deep learning, machine learning, and data analytics software) is set to play a transformative role in the future of medicine, and this Special Collection on “Software as Medical Devices for Cancer Applications” will be a timely contribution to the state-of-the-art with respect to the single most burdensome group of diseases, viz. Cancers. Image credit: [Diabetesmagazijn.nl (Unsplash)](https://unsplash.com/photos/silver-and-white-cross-pendant-necklace-z03Q6GAkqKM). Image credit: [Markus Spiske (Unsplash)](https://unsplash.com/photos/colorful-software-or-web-code-on-a-computer-monitor-Skf7HxARcoc). Software as Medical Devices for Cancer Applications With the explosion of data in medicine, the development of computational methods and workflows to effectively harness this resource and turn it into actionable information is the need of the hour. Artificial Intelligence (including deep learning, machine learning, and data analytics) is set to play a revolutionary role in the future of medicine, and embracing the AI revolution is vital to transform health outcomes. At the same time, attention needs to be paid to ensure that the deployment of AI devices does not worsen healthcare disparities. Democratic precision medicine constitutes the promise of the AI revolution. We have initiated a Special Collection on “Software as Medical Devices for Cancer Applications” that aims to collate cutting-edge research articles and state-of-the-art reviews that push the boundaries to realize the promise of AI for human health. Specifically, this Special Collection calls for contributions that advance software as medical device applications for the screening, early detection, differential diagnosis, prognosis and improved clinical management of cancers. The seminar will provide a technical background for the Special Collection and also examine the ethical dimensions that need to be evaluated prior to the deployment of customized AI physician assistants. Q&A Session with Dr. Vaithilingam Natarajan: SaMD in Clinical Practice Dr Vaithi will be providing a practitioner's perspective on the use of AI in the clinic. The session will cover the state-of-the-art, and the promise of software as medical devices to aid medical/clinical decision making. He will discuss the pitfalls to watch out for in the adoption of such systems, and share ideas on the role and increasing adoption of AI in medicine in the future. This session will be in the form of Q A, and Dr Vaithi will be available afterward to answer questions from the audience.

Two major study topics have emerged because of the challenges in software architecture and ML working together, as modern software systems produce a vast amount of data that is supported particularly by machine learning (ML), and artificial intelligence (AI) to produce useful insights. Software architecture for machine learning systems that primarily concerned with creating architectural methods for creating ML systems more effectively; ii) ML for Software architectures is concerned with creating ML methods for better-developing software systems. This study focuses on the ML-based software systems' architecture to highlight the many architectural methods currently in use. To more clearly identify a set of acceptable standards for designing ML-based software systems, we explore four crucial components of software architecture in this work that demand the focus of ML and software developers. These areas are based on an ML-based software system for addressing challenges in the COVID-19 detecting system.

Continual learning in medical devices: FDA's action plan and beyond

Context. Software systems play a pervasive role in people's lives. Since ethics in software systems is hardly ever considered as a first-class entity, its support of, or hindrance to, values of users and society is usually neglected. Addressing ethics requires both to consider the context in which a software system is used, and the fact that ethical values (e.g., social, cultural and personal) are intrinsically ambiguous and subjective. Software architecture and architecture design decision making are the right levels of abstraction to both consider such a context, and support reasoning about such subjectivities. Goal. We aim to embed ethical aspects in software architecture design decision making. We have studied the state of the art in ethics for software engineering, to identify the ethical aspects influencing software architecture design. We will use the learned ethical aspects to (i) support architecture design decision making with respect to ethics through proposing methods, models, and techniques, and (ii) conduct empirical studies to evaluate them in industrial practice. Method. We will adopt a design science approach for our research which includes three phases: (i) problem understanding, (ii) solution design, and (iii) solution validation. Results. Based on our previous literature study, we build models (and techniques) embedding the ethical aspects to be used in architecture design decision making. Also, we will apply design science research to use and validate these models in a method assisting practitioners and architects to embed ethical aspects in software architecture design.

Machine learning based success prediction for crowdsourcing software projects

WHO set an ambitious vision of malaria control which includes reducing malaria case incidence by 90% by 2030. Many tools and approaches have been considered to enable the progress toward malaria vision. The use of portable smartphones and machine learning (ML) software is a promising one among them. Recently, many ML models have been proposed for malaria detection. From the interview with health workers in the field, we realized ML model should be continuously improved to provide higher accuracy and/or more capability to cover practical issues found in a real setting such as malaria mosquitoes with drug resistance. In this paper, we propose the mobile application for malaria detection which upgrades ML model on its own without depending on internet connection. Unavailability of internet connection is commonly observed in malaria epidemic countries. We also learned that ML model should be not only accurate but also resource-efficient. This motivated us to set up performance metrics for ML model. Based on the metrics, we chose the optimal ML model of Resnet-50. While most of the prior art ML models were optimized in terms of accuracy only, the optimal model of our choice satisfies both accuracy and resource efficiency. With adopting the model, we architect self-upgrading malaria-detection application and it is validated using ATAM (Architecture Tradeoff Analysis Method) to ensure the application works in the resource-constrained setting as desired. Lastly, we develop the prototype application and show it diagnoses malaria parasites as expected. To collect more blood samples and feedback from prospect users, we plan to do testing at local clinics in India and Myanmar.

Code smell detection is essential to improve software quality, enhancing software maintainability, and decrease the risk of faults and failures in the software system. In this paper, we proposed a code smell prediction approach based on machine learning techniques and software metrics. The local interpretable model-agnostic explanations (LIME) algorithm was further used to explain the machine learning model’s predictions and interpretability. The datasets obtained from Fontana et al. were reformed and used to build binary-label and multi-label datasets. The results of 10-fold cross-validation show that the performance of tree-based algorithms (mainly Random Forest) is higher compared with kernel-based and network-based algorithms. The genetic algorithm based feature selection methods enhance the accuracy of these machine learning algorithms by selecting the most relevant features in each dataset. Moreover, the parameter optimization techniques based on the grid search algorithm significantly enhance the accuracy of all these algorithms. Finally, machine learning techniques have high potential in predicting the code smells, which contribute to detect these smells and enhance the software’s quality.

Background: Precision nutrition is based on the integration of individual’s phenotypical and biological characteristics including genetic variants, epigenetic marks, gut microbiota profiles, and metabolite fingerprints as well as medical history, lifestyle practices, and environmental and cultural factors. Thus, nutriomics areas including nutrigenomics, nutrigenetics, nutriepigenetics, nutrimetabolomics, and nutrimetagenomics have emerged to comprehensively understand the complex interactions between nutrients, diet, and the human body’s molecular processes through precision nutrition. Summary: This document from the Ibero-American Network of Nutriomics and Precision Nutrition (RINN22; https://rinn22.com/) provides a comprehensive overview of the concepts of precision nutrition approaches to guide their application in clinical and public health as well as establish the position of RINN22 regarding the current and future state of precision nutrition. Key Messages: The progress and participation of nutriomics to precision nutrition is an essential pillar for addressing diet-related diseases and developing innovative managing strategies, which will be promoted by advances in bioinformatics, machine learning, and integrative software, as well as the description of specific novel biomarkers. In this context, synthesizing and critically evaluating the latest developments, potential applications, and future needs in the field of nutrition is necessary with a holistic perspective, incorporating progress in omics technologies aimed at precision nutrition interventions. This approach must address and confront healthy, social, food security, physically active lifestyle, sanitation, and sustainability challenges with preventive, participatory, and predictive strategies of personalized, population, and planetary nutrition for a precision tailored health. Background: Precision nutrition is based on the integration of individual’s phenotypical and biological characteristics including genetic variants, epigenetic marks, gut microbiota profiles, and metabolite fingerprints as well as medical history, lifestyle practices, and environmental and cultural factors. Thus, nutriomics areas including nutrigenomics, nutrigenetics, nutriepigenetics, nutrimetabolomics, and nutrimetagenomics have emerged to comprehensively understand the complex interactions between nutrients, diet, and the human body’s molecular processes through precision nutrition. Summary: This document from the Ibero-American Network of Nutriomics and Precision Nutrition (RINN22; https://rinn22.com/) provides a comprehensive overview of the concepts of precision nutrition approaches to guide their application in clinical and public health as well as establish the position of RINN22 regarding the current and future state of precision nutrition. Key Messages: The progress and participation of nutriomics to precision nutrition is an essential pillar for addressing diet-related diseases and developing innovative managing strategies, which will be promoted by advances in bioinformatics, machine learning, and integrative software, as well as the description of specific novel biomarkers. In this context, synthesizing and critically evaluating the latest developments, potential applications, and future needs in the field of nutrition is necessary with a holistic perspective, incorporating progress in omics technologies aimed at precision nutrition interventions. This approach must address and confront healthy, social, food security, physically active lifestyle, sanitation, and sustainability challenges with preventive, participatory, and predictive strategies of personalized, population, and planetary nutrition for a precision tailored health.

In agile software development, product backlog items (PBI) are used to capture the user requirements prior to the product implementation. Many types of requirements can be observed within a software project. Proper classification of PBI can positively impact the software development process. PBI can be classified into three categories: user stories, foundational stories, and spikes. After the extreme literature survey, no research was held on classifying the PBI into the categories mentioned above. This paper proposed a machine learning (ML) based approach to classify the PBI into three categories. 4,721 PBI were collected from different software projects and manually labelled into the three classes mentioned above. Then the PBI were cleaned using different pre-processing techniques. Classification models were constructed using ML techniques. The performance of each ML model was evaluated using accuracy, precision, recall, and F1 score. Support vector machine (SVM) outperformed other ML models by providing 88% accuracy.

Modern software systems are increasingly including machine learning (ML) as an integral component. However, we do not yet understand the difficulties faced by software developers when learning about ML libraries and using them within their systems. To fill that gap this thesis reports on a detailed (manual) examination of 3,243 highly-rated Q&A posts related to ten ML libraries, namely Tensorflow, Keras, scikitlearn, Weka, Caffe, Theano, MLlib, Torch, Mahout, and H2O, on Stack Overflow, a popular online technical Q&A forum. Our findings reveal the urgent need for software engineering (SE) research in this area. The second part of the thesis particularly focuses on understanding the Deep Neural Network (DNN) bug characteristics. We study 2,716 high-quality posts from Stack Overflow and 500 bug fix commits from Github about five popular deep learning libraries Caffe, Keras, Tensorflow, Theano, and Torch to understand the types of bugs, their root causes and impacts, bug-prone stage of deep learning pipeline as well as whether there are some common antipatterns found in this buggy software. While exploring the bug characteristics, our findings imply that repairing software that uses DNNs is one such unmistakable SE need where automated tools could be beneficial; however, we do not fully understand challenges to repairing and patterns that are utilized when manually repairing DNNs. So, the third part of this thesis presents a comprehensive study of bug fix patterns to address these questions. We have studied 415 repairs from Stack Overflow and 555 repairs from Github for five popular deep learning libraries Caffe, Keras, Tensorflow, Theano, and Torch to understand challenges in repairs and bug repair patterns. Our key findings reveal that DNN bug fix patterns are distinctive compared to traditional bug fix patterns and the most common bug fix patterns are fixing data dimension and neural network connectivity. Finally, we propose an automatic technique to detect ML Application Programming Interface (API) misuses. We started with an empirical study to understand ML API misuses. Our study shows that ML API misuse is prevalent and distinct compared to non-ML API misuses. Inspired by these findings, we contributed Amimla (Api Misuse In Machine Learning Apis) an approach and a tool for ML API misuse detection. Amimla relies on several technical innovations. First, we proposed an abstract representation of ML pipelines to use in misuse detection. Second, we proposed an abstract representation of neural networks for deep learning related APIs. Third, we have developed a representation strategy for constraints on ML APIs. Finally, we have developed a misuse detection strategy for both single and multi-APIs. Our experimental evaluation shows that Amimla achieves a high average accuracy of ∼80% on two benchmarks of misuses from Stack Overflow and Github.

Today, a greater generation of information is produced as a consequence of the technological development of society. The Internet has facilitated the access and extraction of this information, thus pursuing the automatic discovery of the knowledge contained within. In this context, data mining aims to discover patterns, profiles and trends of a large volume of data, for which multiple learning techniques are available. The selection of which technique to use depends on the type of result desired to obtain and the data that are available, considering that the algorithms for these tasks date mostly from the early twentieth century and are now the basis of these new technologies. The aim of this study is to show the development of these techniques in the field of scientific research and to present the evolution of algorithms and software for data mining in recent years. To this end, the systematic literature review methodology was applied, as it is considered a systematic process that identifies, evaluates, and interprets the work of researchers in a chosen field. As a result, we present a comparative analysis of the most outstanding software: Alteryx, TIBCO Data Science, RapidMiner and WEKA, their capacities for data mining processes and a description of the algorithms and techniques of machine learning that are currently on the rise.

Software architecture involves a series of decisions based on many factors in a wide range of software development. Architects face recurring issues in different software architecture design, and to reduce huge cost and risks, software architecture decisions can rely on a set of idiomatic patterns commonly named architectural styles or patterns. Architectural pattern determines the vocabulary of components and connectors that are used in instances of the pattern together with a set of constraints to combine the two. Little contemporary data exists to document actual practices used by software professionals when selecting and incorporating architectural patterns for their projects in industry. Therefore, a comprehensive survey of software professionals was conducted to attempt to discover these practices. This exploratory survey and its quantitative results offer opportunities for further interpretation and comparison. Data from this survey are presented in this paper and include characteristics of projects, practices, organizations, and practitioners related to the usage of architectural patterns. Some of the notable findings include that architectural patterns are widely used in software projects with the Model–View–Controller being the most common. Despite reported difficulties in incorporating architectural patterns, the majority of the software professionals revealed that patterns were the most essential for completing the projects. The most difficult pattern to implement and the most expensive to adopt was the peer-to-peer, while the easiest was the client–server.

Effort estimations are a crucial aspect of software development. The tasks should be completed before the start of any software project. Accurate estimations increase the chances of project success, and inaccurate information can lead to severe issues. This study systematically reviewed the literature on effort-estimating models from 2015-2024, identifying 69 relevant studies from various publications to compile information on various software work estimation models. This review aims to analyze the models proposed in the literature and their classification, the metrics used for accuracy measurement, the leading model that has been chiefly applied for effort estimation, and the benchmark datasets available. The study utilized 542 relevant articles on software development, cost, effort, prediction, estimation, and modelling techniques in the search strategy. After 194 selections, the authors chose 69 articles to understand ML applications in SEE comprehensively. The researchers used a scoring system to assess each study’s responses (from 0 to 5 points) to their research questions. This helped them identify credible studies with higher scores for a comprehensive review aligned with its objectives. The data extraction process identified 91% (63) of 69 studies as either highly or somewhat relevant, demonstrating a successful search strategy for analysis. The literature review on SEE indicates a growing preference for ML-based models in 59% of selected studies. 17% of the studies chosen favor hybrid models to overcome software development challenges. We qualitatively analyzed all the literature on software effort estimation using expert judgment, formal estimation techniques, ML-based techniques, and hybrid techniques. We discovered that researchers have frequently used ML-based models to estimate software effort and are currently in the lead. This study also explores the application of feature importance and selection in machine learning models for Software Effort Estimation (SEE) using popular algorithms like support Vector Machine (SVM), AdaBoost (AB), Gradient Boost (GB), and Random Forest (RF) with six benchmark datasets like CHINA, COCOMO-NASA2, COCOMO, COCOMO81, DESHARNAIS, and KITCHENHAM. We analyze the dataset descriptions and feature importance of the dataset analysis using ML models for choosing crucial play attributes in SEE.

Playing a pivotal role in software development, the Constructive Cost Model (COCOMO) offers a systematic and structured approach to cost estimation. It stands as a widely utilized model, aiding project managers in estimating the required effort, time, and cost for software development projects. COCOMO takes into consideration diverse factors, including the project's size, complexity, and the experience of the development team. The utilization of COCOMO empowers software development teams to make informed decisions related to resource allocation, project scheduling, and budgeting. Its application extends to managing expectations, enhancing project planning, and mitigating the risk of cost overruns. The integration of machine learning assumes a critical role in advancing cost estimation within the realm of software development, specifically through COCOMO. Through the utilization of machine learning algorithms, COCOMO gains the capability to analyze and interpret extensive datasets, taking into account numerous complex factors that influence project costs. This work aims to propose a model for software cost estimation using an advanced machine learning technique, i.e. Adaptive boosting, which has improved accuracy, reduced overfitting, effectiveness with imbalanced data, and good generalization capabilities. The proposed work may contribute to the success of software projects, providing a reliable and comprehensive framework for cost estimation.