Fault concealment in complex software programs and the difficulty of generating test cases to detect such faults present significant challenges in software testing. To resolve these challenges, this paper suggests a novel method that integrates mutation testing, fuzzy clustering, convolutional neural networks (CNN), and particle swarm optimization (PSO) to efficiently generate test cases that cover multiple paths with numerous faults (mutant branches). Initially, mutation-based paths are classified using fuzzy clustering based on their coverage difficulty and similarity. A multi-feature CNN model (MF_CNNpro) is then constructed and trained on the paths of each cluster. Finally, the predicted particles from the MF_CNNpro model are used as the initial population for PSO, which evolves to generate the test cases. The proposed method is evaluated on six test programs, and the results demonstrate that it significantly improves clustering separation and reduces clustering compactness. By selecting only the cluster center paths to construct the MF_CNNpro model, training and prediction costs are effectively reduced. Moreover, the use of MF_CNNpro and PSO to select representative individuals as the initial population greatly enhances the evolutionary efficiency of PSO. The proposed method outperforms traditional approaches in clustering, prediction, and test data generation. Specifically, the SC clustering method improves cluster separation (SP) by 0.021, reduces compactness (CP) by 0.054, and decreases clustering rate (CR) by 4.97%, thereby enhancing clustering precision. The MF_CNNpro model improves the IA metric by 38.2% and reduces the U-Statistic and MSE by 83.0% and 97.9%, respectively, optimizing prediction performance. The MF_CNNpro+PPSOpro method increases the path coverage success rate from 47.9% to 97.4% (a 103.3% improvement), reduces the number of iterations by 84.1%, and decreases execution time by 95.6%, significantly improving generation efficiency.

With the exponential growth of digital transactions, organizations across banking, fintech, e-commerce, and telecommunications face increasingly sophisticated fraud attempts. Traditional fraud detection systems, primarily rule-based and manually configured, struggle to keep pace with evolving fraud patterns and exhibit high false-positive rates. Artificial Intelligence (AI), particularly machine learning (ML) and deep learning (DL), has emerged as a transformative solution by enabling pattern recognition, anomaly detection, behavioral analytics, and real-time decisioning at scale. This paper provides a structured overview of AI-driven fraud detection models, their technical components, data pipelines, deployment architectures, and evaluation frameworks. It compares traditional rule-based approaches with supervised, unsupervised, and hybrid AI methods, and discusses practical challenges such as class imbalance, concept drift, data quality, and latency constraints in real-time payment environments. The paper also highlights explainability challenges, regulatory implications under frameworks such as GDPR and PSD2, and future innovations including federated learning, graph neural networks, and generative AI for adversarial testing and synthetic data generation. Experimental discussion and case-style examples from card-not-present, account takeover, and telecom subscription fraud scenarios illustrate how AI can significantly improve fraud detection accuracy and operational efficiency while emphasizing that careful governance, model monitoring, and responsible AI practices are essential for trustworthy deployment.

Optimizing test data generation using SI_CNNpro-enhanced MGA for mutation testing

Image Formation in the Environment of a Mobile Robot''''s Choice and Their Classification Using Neural Networks and Logical-Linguistic Classification Algorithms. The article investigates the task of improving the accuracy and speed of image classification for mobile robot and UAV control systems under conditions of data uncertainty. An integrated approach combining fuzzy logic methods (logical-linguistic classification (LLC)) and neural network technologies is proposed. A mathematical model for image representation using attribute membership functions has been developed, allowing it to work with noisy and incomplete data. An algorithm for generating test data based on etalon images with an adjustable noise level (0 − 100%) was created. A comparative testing of the neural network approach and the LLC algorithm was conducted on sample sizes ranging from 680 to 68 000 images. It was experimentally established that the neural network demonstrates high efficiency with large data volumes and high noise levels (> 80%), while the LLC algorithm is more effective with small samples and moderate noise levels (50 − 60%). The minimum training sample size for stable operation of the neural network was determined to be 6 800 images. The practical significance of the work lies in the development of an adaptive classification system capable of operating in real-world conditions of robotic complexes with variable levels of informational uncertainty.

<div>This article suggests a validation methodology for autonomous driving. The goal is to validate front camera sensors in advanced driver-assist systems (ADAS) based on virtually generated scenarios. The outcome is the CARLA-based hardware-in-the-loop (HIL) simulation environment (CHASE). It allows the rapid prototyping and validation of the ADAS software. We tested this general approach on a specific experimental application/setup for a vehicle front camera sensor. The setup results were then proven to be comparable to real-world sensor performance. The CARLA simulation environment was used in tandem with a vehicle CAN bus interface. This introduced a significantly improved realism to user-defined test scenarios and their results. The approach benefits from almost unlimited variability of traffic scenarios and the cost-efficient generation of massive testing data.</div>

Conversational AI systems are now part of various industries, requiring efficient testing methods to guarantee reliability, accuracy, and user satisfaction at production levels. Intelligent test data generation has become a vital part of developing and assessing these systems, counteracting the core issues present due to the nature of natural language as well as changing user interactions. This in-depth survey explores existing methodologies for creating efficient test datasets that mimic actual conversations and corner cases with the help of cutting-edge machine learning, natural language processing, and automation technologies. The shift from basic rule-based chatbots to high-end neural dialogue systems has revolutionized the testing arena with the need for systems capable of dealing with contextual comprehension, multi-turn dialogue, emotional undertones, and specialized domain vocabulary across a variety of languages and cultural backgrounds. Classical software testing practices are found wanting for the probabilistic and context-based nature of conversational AI, resulting in enormous system validation gaps. The review delves into different generation strategies involving rule-based approaches, data augmentation methods, generative models, adversarial testing, and user simulation platforms. Modern quality assurance issues include semantic coherence verification, pragmatic appropriateness assessment, cultural sensitivity validation, scalability needs, domain adaptation challenges, and privacy issues. Future directions place focus on human-in-the-loop integration, context-sensitive generation abilities, cross-lingual and multimodal data generation, and ongoing testing frameworks that evolve according to changing system capabilities.

ABSTRACT Background High‐level system testing of applications that use data from e‐Government services as input requires test data that is real‐life‐like but where the privacy of personal information is guaranteed. Applications with such strong requirement include information exchange between countries, medicine, banking, and so on. This review aims to synthesise the current state‐of‐the‐practice in this domain. Objectives The objective of this Systematic Review is to identify existing approaches for creating and evolving synthetic test data without using real‐life raw data. Methods We followed well‐known methodologies for conducting systematic literature reviews, including the ones from Kitchenham and PRISMA as well as guidelines for analysing the limitations of our review and its threats to validity. Results A variety of methods and tools exist for creating privacy‐preserving test data. Our search found 1013 publications in IEEE Xplore, ACM Digital Library, and SCOPUS. We extracted data from 75 of those publications and identified 37 approaches that answer our research question partly. A common prerequisite for using these methods and tools is direct access to real‐life data for data anonymization or synthetic test data generation. Nine existing synthetic test data generation approaches were identified that were closest to answering our research question. Nevertheless, further work would be needed to add the ability to evolve synthetic test data to the existing approaches. Conclusions None of the publications covered our requirements completely, only partially. Synthetic test data evolution is a field that has not received much attention from researchers but needs to be explored in Digital Government Solutions, especially since new legal regulations are being put in force in many countries.

The RTgill-W1 in vitro assay is a new approach method designed as an alternative to one of the most widely used toxicity tests globally, the fish acute lethality test. The RTgill-W1 assay is standardized (Organisation for Economic Co-operation and Development; International Organisation of Standardization), but the test conditions could be optimized to allow for higher throughput, better replication, and lower costs. This study explores potential optimizations regarding the culturing conditions, plate format, and reference toxicant testing to make the RTgill-W1 assay more practical for widespread implementation. We demonstrate that the RTgill-W1 culture can be routinely split 1:3 without impacting test sensitivity (p = 0.207 to 0.612), which allows all work to be contained to a standard 5-day work week and 1.3× more tests over the current test methods. The test can adopt a 96-well plate format without impacting sensitivity (p = 0.672 to 0.889), dramatically improving the replication of the exposure wells and test controls and generating test data from a single plate. The fluorescent signal does not bleed across the smaller wells in the 96-well format to impact test endpoints (p = 0.465 to >0.999). The reference toxicity test concentrations can be modified to increase confidence in the point estimate (median effect concentration), allowing for more effective monitoring of assay performance. These optimizations improve the practicality and decrease the costs of the RTgill-W1 assay, which is particularly desirable for implementation in commercial and government laboratories that conduct regulatory toxicity testing.

Cold spray is an advanced additive manufacturing technology that is capable of restoration of damaged metallic components without exposing them to high temperatures. To expand the use of cold spray from restoring the geometry and structure of defective parts to full remanufacturing, extending their lifespan beyond the original life cycle by replacing internally damaged areas (typically only 10–15% of the part’s volume), the first step is to accurately assess the damage at the part’s hot spot. This study explores the capabilities of X-ray diffraction (XRD) as a non-destructive testing method for assessing damage in AA6061-T6. A set of dog-bone samples was prepared to introduce controlled damage at different levels. X-ray diffraction measurements were conducted on these samples to generate test data, to assess dislocation densities. These values offer a quantified measure of internal damage and provide insight into the microstructural evolution under fatigue loading. By using this method, this study aims to develop a reliable method for pre-additive remanufacturing® damage assessment. In corroboration with earlier studies, we show that XRD can effectively detect internal material damage using dislocation densities through XRD-measured parameters such as full width at half maxima (FWHM), a measure of XRD peak broadening used for analyzing dislocation and strain. Integrating XRD-based damage assessment with cold spray additive manufacturing can enable precise and localized repairs. By implementing cold spray remanufacturing, this method can significantly reduce material waste, a major contributor to the greenhouse gas emissions, and extend components' lifespans across various industries, promoting sustainability and circular economy.

This paper presents an intelligent AI test modeling framework for computer vision systems, focused on image-based systems. A three-dimensional (3D) model using decision tables enables model-based function testing, automated test data generation, and comprehensive coverage analysis. A case study using the Seek by iNaturalist application demonstrates the framework’s applicability to real-world CV tasks. It effectively identifies species and non-species under varying image conditions such as distance, blur, brightness, and grayscale. This study contributes a structured methodology that advances our academic understanding of model-based CV testing while offering practical tools for improving the robustness and reliability of AI-driven vision applications.

The Firefly Algorithm (FA), while effective for complex optimization, suffers from inherent limitations such as search oscillation and low convergence precision. To address these issues, a firefly algorithm with multiple learning ability based on gender difference (MLFA-GD) is proposed. Firstly, the algorithm evenly divides the randomly initialized population into male and female subgroups. Then a male firefly learning strategy which incorporated a partial attraction model combining with an escape mechanism, and a female firefly learning strategy guided by both the generalized centroid of the male subgroup and the global optimal individual are designed separately. Additionally, a random walk strategy is further incorporated to refine the optimization accuracy. Different from existing gender-based FA variants, male fireflies either fly toward brighter female fireflies or move away from weaker individuals to enhance exploration capability. Meanwhile, female fireflies update positions guided by two elite male individuals, effectively leveraging historical search information to improve exploitation capability. The performance is evaluated on 23 numerical functions, 30 CEC 2017 benchmark functions and an automatic test data generation problem. The experiment comparison results with six FA variants and ten popular meta heuristic algorithms confirm its enhanced search capability and significantly higher optimization precision, validating its effectiveness in balancing exploration and exploitation.

Efficient path coverage-based test data generation using an enhanced pelican algorithm

Application Programming Interfaces (APIs) constitute the backbone of modern software architectures, enabling seamless communication and integration of heterogeneous systems. It is crucial to guarantee the reliability and correctness of such interactions, making API contract testing an essential discipline. However, traditional API contract testing approaches have several significant limitations, including manual effort to write test cases, difficulty in keeping the test suites in sync with evolving API specifications, challenges in attaining complete test coverage and creating practical, varied test data. This paper gives a detailed description of how Artificial Intelligence (AI) may be leveraged to overcome these restrictions and significantly enhance the efficiency of test generation in API contract testing. We cover several AI techniques, including machine learning-based approaches to learn from API specs, historical data, and network traffic; the revolutionary potential of Generative AI and Large Language Models (LLMs) to synthesize automated test cases from natural language descriptions and synthetic test data generation; and reinforcement learning-based test optimization. The survey examines the specific benefits of AI, such as automated contract checking, intelligent test data generation, and the possibility of self-healing tests that adapt to API changes. We also cover the significant challenges and determinants concerning AI adoption in this space, including data quality, explainability, integration problems, and the necessity of human-in-the-loop verification. By consolidating results of current research and advances, this paper aims to provide a structured understanding of the state of the art, identify possible future directions, and outline the ethical consequences of applying AI to achieve more efficient, robust, and scalable API contract testing. Keywords: API Contract Testing, Test Generation, Artificial Intelligence (AI), Machine Learning (ML), Generative AI, Automated Test Case Generation.

This research investigates the efficacy and challenges of employing generative Artificial Intelligence (AI) models—specifically Anthropic Claude, OpenAI GPT, and Google Gemini, orchestrated via the Cline extension in Visual Studio Code—to construct a .NET MAUI test application for validating FinTech Software Development Kits (SDKs). The study focuses on automating the generation of XAML for user interfaces and C# for backend logic, targeting critical FinTech workflows such as wallet provisioning on Android and iOS platforms. The methodology involved an iterative AI-assisted development process, encompassing AI-driven planning, code generation, extensive human-led refinement, and rigorous testing using a mock SDK and Appium for UI automation. Hypothesized results suggested significant acceleration in initial boilerplate code generation, though a substantial portion (40-60% for XAML, 30-50% for C#) required manual rework to address framework-specific nuances, ensure code quality, and implement robust error handling. Key challenges identified include the AI's inconsistent understanding of .NET MAUI's XAML dialect, limitations in managing complex UI state, occasional AI hallucinations, and the need for highly specific, context-rich prompts. Despite these hurdles, the final human-refined test application successfully automated the validation of the designated FinTech workflow across both platforms. The findings indicate that while AI serves as a powerful accelerator in test application development, expert human oversight remains indispensable for ensuring the quality, security, and framework compliance of the generated code. The study concludes that AI dramatically reshapes the role of the test automation engineer towards that of an AI orchestrator and critical validator, and outlines future research directions including the development of fine-tuned AI models for specific frameworks like .NET MAUI and AI-driven test data generation for complex FinTech scenarios.

The decision tree test method works as a flowchart structure for conversational flow. It has predetermined questions and answers that guide the user through specific tasks. Inspired by principles of the decision tree test method in software engineering, this paper discusses intelligent AI test modeling chat systems, including basic concepts, quality validation, test generation and augmentation, testing scopes, approaches, and needs. The paper’s novelty lies in an intelligent AI test modeling chatbot system built and implemented based on an innovative 3-dimensional AI test model for AI-powered functions in intelligent mobile apps to support model-based AI function testing, test data generation, and adequate test coverage result analysis. As a result, a case study is provided using a mental health and emotional intelligence chatbot system, Wysa. It helps in tracking and analyzing mood and helps in sentiment analysis.

In operational testing contexts, testers face dual challenges of constrained timeframes and limited resources, both of which impede the generation of reliability test data. To address this issue, integrating data from similar systems with test data can effectively expand data sources. This study proposes a systematic approach wherein the mission of the system under test (SUT) is decomposed to identify candidate subsystems for data combination. A phylogenetic tree representation is constructed for subsystem analysis and subsequently mapped to a mixed-integer programming (MIP) model, enabling efficient computation of similarity factors. A reliability assessment model that combines data from similar subsystems is established. The similarity factor is regarded as a covariate, and the regression relationship between it and the subsystem failure-time distribution is established. The joint posterior distribution of regression coefficients is derived using Bayesian theory, which are then sampled via the No-U-Turn Sampler (NUTS) algorithm to obtain reliability estimates. Numerical case studies demonstrate that the proposed method outperforms existing approaches, yielding more robust similarity factors and higher accuracy in reliability assessments.

The escalating complexity of mobile application testing, driven by device fragmentation and rapid development cycles, necessitates advanced solutions for ensuring software quality. Mobile device farms provide essential infrastructure for testing across diverse real-world devices, while test automation accelerates repetitive validation tasks. However, significant manual effort persists in test design, data preparation, script maintenance, and results analysis. This research investigates the integration of Generative Artificial Intelligence (GenAI) within mobile device farms to address these challenges and enhance mobile test automation. Key GenAI applications explored include the automated generation of diverse and realistic test data, the creation of test scripts from natural language or user flows, the simulation of complex user interactions and edge cases, the intelligent analysis of test results and logs for anomaly detection and root cause analysis, and the potential optimization of device allocation and test scheduling within the farm. Employing a methodology based on literature review and conceptual framework analysis, this paper examines potential methodologies, frameworks, algorithms, and tools for implementing GenAI solutions in this context. The analysis highlights potential benefits such as improved test coverage, increased efficiency, reduced manual effort, faster feedback cycles, and enhanced defect detection capabilities. Concurrently, it critically assesses significant challenges, including implementation complexity, data privacy and security concerns, the reliability and accuracy of generated artifacts, integration difficulties, and computational costs. The findings suggest that GenAI holds considerable potential to transform mobile testing within device farms, shifting towards a more intelligent, adaptive, and efficient paradigm, although its role is likely to be that of a powerful assistant augmenting human expertise rather than a complete replacem

Adherence to clinical guidelines supports high quality patient care. Conformance checking, a feature of process mining, can potentially automate the assessment of adherence to clinical guidelines in practice. This paper investigates appropriate conformance checking in practice. Conformance checking in practice was simulated with generated test data, a FHIR server and process mining tools. A corresponding literature review was conducted in parallel. Activities of clinical guidelines or in healthcare processes should be coded using clinical nomenclature to support conformance checking. SNOMED CT should be used as a nomenclature and activities should be coded with SNOMED concepts of the type "procedure".

Automatic near-optimal generation of software test data for critical paths

Over the past decade, there has been a significant surge in interest regarding the application of machine learning (ML) across various tasks. Due to this interest, the adoption of ML-based systems has gone mainstream. It turns out that it is imperative to conduct thorough software testing on these systems to ensure that they behave as expected. However, ML-based systems present unique challenges for software testers who are striving to enhance the quality and reliability of these solutions. To cope with these testing challenges, we propose novel test adequacy criteria centered on decision tree models. Our criteria diverge from the conventional method of manually collecting and labeling data. Instead, our criteria relies on the inherent structure of decision tree models to inform the selection of test inputs. Specifically, we introduce decision tree coverage (DTC) and boundary value analysis (BVA) as approaches to systematically guide the creation of effective test data that exercises key structural elements of a given decision tree model. Additionally, we also propose a mutation based criterion to support the validation of ML-based systems. Essentially, this approach involves applying mutation analysis to the decision tree structure. The resulting mutated trees are then used as a reference for selecting test data that can effectively identify incorrect classifications in ML models. To evaluate these criteria, we carried out an experiment using 16 datasets. We measured the effectiveness of test inputs in terms of the difference in model’s behavior between the test input and the training data. According to the results of the experiment, our criteria can be used to improve the test data selection for ML applications by guiding the generation of diversified test data that negatively impact the prediction performance of models.