Correlated quality metrics extracted from a source code repository can be utilized to design a model to automatically predict defects in a software system. It is obvious that the extracted metrics will result in a highly unbalanced data, since the number of defects in a good quality software system should be far less than the number of normal instances. It is also a fact that the selection of the best discriminating features significantly improves the robustness and accuracy of a prediction model. Therefore, the contribution of this paper is twofold, first it selects the best discriminating features that help in accurately predicting a defect in a software component. Secondly, a cost-sensitive logistic regression and decision tree ensemble-based prediction models are applied to the best discriminating features for precisely predicting a defect in a software component. The proposed models are compared with the most recent schemes in the literature in terms of accuracy, area under the curve, and recall. The models are evaluated using 11 datasets and it is evident from the results and analysis that the performance of the proposed prediction models outperforms the schemes in the literature.

With the increasing adoption of Internet of Medical Things (IoMT) devices, modern healthcare systems face persistent challenges related to data privacy, device heterogeneity, communication overhead, and scalability especially in consumer electronics environments. This paper hypothesizes that a hierarchical and privacy-preserving federated learning architecture can address these challenges and enable efficient, secure, and personalized healthcare insights across distributed IoMT networks. To validate this, we present the Adaptive Federated Learning Framework (AFLF), a novel framework tailored for resource-constrained consumer devices such as wearable ECG monitors, glucose trackers, and portable neuro-sensors. AFLF incorporates four core components: (1) a hierarchical edge–fog–cloud model for scalable and latency-aware training, (2) a Secure Data Collaboration Protocol (SDCP) that combines blockchain-based audit trails with differential privacy to ensure integrity and confidentiality, (3) the Adaptive Personalized Federated Learning Algorithm (APFLA) that dynamically tunes learning rates and model weights for personalized inference, and (4) a ternary threshold-based gradient compression technique that reduces communication overhead by 45.3%. The framework is deployed on Raspberry Pi 4 (edge) and NVIDIA Jetson AGX Xavier (fog) platforms, using PyTorch Mobile for implementation. Experimental validation using four real-world datasets, CardioFit (ECG), GlucoWatch (glucose), PulseOx (PPG), and NeuroMotion (EEG), demonstrates that AFLF improves model accuracy by up to 12% reduces training time by 38% and preserves user privacy with a differential privacy budget of ε = 2.1. These results confirm that AFLF offers a robust and scalable privacy-preserving federated learning solution for next-generation consumer-centric smart healthcare applications.

The manipulation of digital images has become a popular trend. Due to the development of image processing tools and visualization techniques integrating deep learning and artificial intelligence (AI) algorithms (in particular generative adversarial networks–GAN), this can pose serious threats to privacy and security. In recent years, Deepfake algorithms have been designed to exchange faces or modify facial features, potentially leading to more severe problems in this context. In this manuscript, we provide a comprehensive review of the two most important facial image processing technologies: (i) deepfake face manipulation; and (ii) face manipulation detection techniques. Furthermore, we explore the state-of-the-art of popular GAN techniques. In particular, three types of Deepfake face detection technologies are reviewed: (i) hand-crafted features (ii) artifacts; and (iii) learning-based features, while highlighting related improvements and challenges. Furthermore, this article discusses potential challenges and promising research directions for future investigation. We believe that this review has been organized to provide a structured analysis of important research papers and to discuss each study’s main findings and conclusions. Our investigation reveals shortcomings in manipulation detection benchmarks due to real-world scenario variations and biased dataset comparisons. Current research priorities revolve around enhancing GAN training stability, resolution, and manipulable facial features. Moreover, GANs have shown superior results in identifying fake images; however, their reliance prompts a systematic approach to detecting fakes. This dependency raises questions about detecting fake images with or without manipulated GAN architecture, urging the need for novel computational techniques to identify manipulations without GAN assistance.

Real-world business processes are dynamic, with event logs that are generally unstructured and contain heterogeneous business classes. Process mining techniques derive useful knowledge from such logs but translating them into simplified and logical segments is crucial. Complexity is increased when dealing with business processes with a large number of events with no outcome labels. Techniques such as trace clustering and event clustering, tend to simplify the complex business logs but the resulting clusters are generally not understandable to the business users as the business aspects of the process are not considered while clustering the process log. In this paper, we provided a multi-stage hierarchical framework for business-logic driven clustering of highly variable process logs with extensively large number of events. Firstly, we introduced a term contrail processes for describing the characteristics of such complex real-world business processes and their logs presenting contrail-like models. Secondly, we proposed an algorithm Novel Hierarchical Clustering (NoHiC) to discover business-logic driven clusters from these contrail processes. For clustering, the raw event log is initially decomposed into high-level business classes, and later feature engineering is performed exclusively based on the business-context features, to support the discovery of meaningful business clusters. We used a hybrid approach which combines rule-based mining technique with a novel form of agglomerative hierarchical clustering for the experiments. A case-study of a CRM process of the UK’s renowned telecommunication firm is presented and the quality of the proposed framework is verified through several measures, such as cluster segregation, classification accuracy, and fitness of the log. We compared NoHiC technique with two trace clustering techniques using two real world process logs. The discovered clusters through NoHiC are found to have improved fitness as compared to the other techniques, and they also hold valuable information about the business context of the process log.