ABSTRACT Background The reliable operation of diesel generators, which are critical devices in many industrial applications in maintaining continuous power supply, is essential. Objective This research presents an advanced Artificial Intelligence of Things (AIoT) fault diagnosis framework that combines metaheuristic optimization and two‐stage deep learning methods to improve the fault classification accuracy of diesel generators. Methods The improved particle swarm optimization (IPSO) is used to fine‐tune the hyperparameters of bidirectional long short‐term memories (BiLSTM) combined with convolutional neural networks (CNN). The developed hybrid model aims to exploit the power of both the sequential learning of BiLSTMs and the feature extraction capability of CNNs to achieve superior fault diagnosis performance. Seven established baselines provide the point of reference for the IPSO‐BiLSTM‐CNN framework: recurrent neural network (RNN), CNN, gated recurrent unit (GRU), LSTM, bidirectional RNN (BiRNN), BiLSTM and CNN‐BiLSTM. Results Experimental results demonstrate that IPSO‐BiLSTM‐CNN methods outperform existing models in terms of accuracy and stability, achieving relative reductions of 71.56% in loss, 73.79% validating loss, 64.63% mean absolute error (MAE), 69.75% validating MAE, 23.17% cosine proximity (CP), 21.99% validating CP, 63.08% mean squared error (MSE), 71.55% validating MSE, together with relative improvements of 4.79% precision (PRE), 3.79% validating PRE, 5.8% recall (REC), and 5.29% validating REC. Conclusions This highlights its potential in practical AIoT applications for diesel generator monitoring and maintenance.
Context Pull-based development model provides a streamlined contribution mechanism that encourages volunteer contributors to participate in Open Source Software (OSS) projects. However, a substantial number of valid Pull Requests (PRs) are abandoned by their contributors, leading to wasted effort for both contributors and maintainers, and negatively impacting the health of OSS projects.Objective This study investigates the association between contributor-abandoned PRs and a range of factors spanning PR, project, review process, and contributor dimensions.Method To achieve this objective, we conducted an empirical analysis of eight factors across four dimensions in ten mature and diverse OSS projects. Specifically, we considered PR-related factors (PR submission day and time, authorship type, and test file inclusion), a project-related factor (number of core maintainers), a review process-related factor (sentiment in discussion comments), and contributor-related factors (contributor experience and contributor gender). We performed statistical analyses to explore the associations between these factors and contributor-abandoned PRs.Results Our results indicate that, at the PR level, the inclusion of test files is significantly associated with lower PR abandonment in certain studied projects. In contrast, submission day, submission time interval, and the number of authors show no statistically significant differences in abandonment. Within the review process, a higher proportion of negative comments and a lower proportion of positive comments are associated with increased abandonment. At the project level, the relationship between the number of core maintainers and abandonment varies across the studied projects. At the contributor level, lower experience is associated with higher abandonment, and female contributors exhibit lower abandonment rates than male contributors.Conclusion These findings highlight the factors that are significantly associated with contributor-abandoned PRs, as well as those that show no notable association, offering valuable insights for the OSS community and researchers.
Objectives Apache Spark is widely used for distributed data processing, but accurately predicting application execution time remains challenging because performance depends on application structure, resource configuration, and executor-allocation behavior. This article presents two deterministic, graph-based simulation models for predicting the execution time of Spark applications under static and dynamic executor allocation.Methods The first model, the Static Allocation Model (SAM), represents an application as a monolithic directed acyclic graph (DAG) of stages and simulates execution using Spark's scheduling behavior under a fixed number of executor cores. To capture more realistic execution environments, this framework is extended to a Dynamic Allocation Model (DAM), which accounts for runtime changes in executor resources using backlog-time and idle-time thresholds. DAM dynamically adjusts executor allocation during execution and integrates these changes within the deterministic scheduling framework.Result Validation of SAM on three TPC-DS benchmark queries-Query-64, Query-26, and Query-52-yielded prediction errors of 2.9%, 4.3%, and 0.4%, respectively, with a mean absolute error of 2.5%. Experimental evaluation of DAM on TPC-DS Query-52 and the SparkBench K-Means workload yielded prediction errors of 5% and 4.7%, respectively.Conclusion To the best of our knowledge, this work presents the first deterministic approach for modeling Spark application execution time under dynamic executor allocation. The proposed models are lightweight, interpretable, and require minimal input data, offering a practical alternative for performance prediction and resource optimization in distributed data-processing systems.
ABSTRACT Context Roughly 20% to 28% of the GDP of the Kashmiri agricultural sector are based on high‐value crops, such as apples, walnuts, saffron, and apricots. These face significant economic challenges due to crop damage caused by the Indian crested porcupine (Hystrix indica). Objective This research aims to study and develop an effective real‐time management solution for wildlife intrusion in orchards, overcoming the limitations of traditional deterrence methods. Methods : We propose a multilayered edge‐IoT architecture that combines motion, acoustic, and visual sensing with edge‐based machine learning and automated wildlife monitoring. The system leverages lightweight YOLOv8n models running on edge devices, secure communication frameworks, and intelligent monitoring with a user‐facing application for alerts and analytics. A comprehensive wildlife detection dataset was developed by integrating multiple public datasets and converting segmentation annotations into object‐detection bounding boxes. The final dataset contains 1758 images with 2249 annotated instances across three classes: porcupine, hedgehog, and others. Object detection is performed using YOLOv8n on resource‐constrained edge devices for efficient real‐time inference. Results Experimental results show that YOLOv8n achieves 93.9% mAP@0.5 and 86.4% mAP@0.5:0.95 with 84.0% precision and 89.9% recall on the test dataset. The porcupine class specifically achieves 99.5% mAP@0.5 and 95.4% mAP@0.5:0.95, demonstrating highly reliable porcupine detection performance. The lightweight YOLOv8n architecture achieves approximately 4.9 ms inference time per image on the validation host, and 81 ms per frame on a Jetson Nano edge node, enabling efficient edge deployment for real‐time orchard monitoring. All reported figures are obtained from real image data combined with simulated multi‐sensor confidence modeling; no result derives from a deployed orchard installation, and the scope of validation is stated explicitly. Conclusion Furthermore, the proposed framework establishes a foundation for future privacy‐preserving and Agentic AI‐enabled intelligent wildlife monitoring systems capable of adaptive decision‐making, autonomous monitoring, and collaborative edge intelligence under dynamic environmental conditions
ABSTRACT Background Application programming interfaces (APIs) help distributed systems and microservices to expose their functionalities and serve as a means for communication among them. Understandability and readability of APIs are important for effectively comprehending and using APIs. Despite the presence of established API design rules and guidelines (i.e., design patterns), many poor design practices (i.e., antipatterns) are prevalent in APIs. APIs with high‐quality design are essential both for API providers and client developers. Goal This paper aims to (1) assess the linguistic design quality of APIs in distributed systems and microservices by automatically detecting antipatterns and design patterns in APIs and (2) evaluate the impact of linguistic patterns and antipatterns on the understandability and readability of APIs. Method We rely on syntactic and semantic analyses for automatic assessment of the design quality of APIs using detection heuristics. Syntactic analysis involves analyzing the structure and syntax of the APIs, while semantic analysis involves analyzing API documentation, descriptions, and parameters. We also conducted a survey and asked participants to determine the purpose of API snippets (comprehension task) and rate the difficulty in the understandability and readability of those API snippets. Findings We achieved an average accuracy of more than 94% in detecting patterns and antipatterns. Our detection results also suggest that antipatterns are prevalent in the APIs of distributed systems and microservices. Our survey findings revealed that adherence to linguistic patterns significantly enhances the understandability and readability of APIs. In contrast, linguistic antipatterns negatively affect the understandability and readability of Web APIs. Conclusion Our findings will assist API developers in identifying poor design practices and improving their APIs' design quality. Findings from our survey also highlight the importance of adhering to good API design practices to enhance the understandability and readability of APIs.
