Deep Q-Network (DQN) is a popular deep reinforcement learning algorithm that has demonstrated promising results across a variety of domains. DQN presents a promising solution to the challenge of lowering false positives in software defect prediction, thereby enhancing the reliability of the prediction performance. In software defect prediction, false positives occur when the prediction model incorrectly predicts code changes to be defective. Consequently, developers waste time and resources on non-existent defects. This paper advocates for employing DQN in software defect prediction, focusing on minimizing false positives and maximizing the prediction performance. Throughout the training phase, the model learns to predict defect-prone following a reward policy aimed at reducing false results. Experimental findings show that the proposed DQN outperforms baseline classifier, improving the prediction accuracy of true defects by up to 27% when using only 20% efforts. The results show that the effectiveness of DQN in tackling false positives, thereby emphasizing the significance of incorporating dynamic reward in predicting software defects.

Hadoop MapReduce reactively detects and recovers faults after they occur based on the static heartbeat detection and the re-execution from scratch techniques. However, these techniques lead to excessive response time penalties and inefficient resource consumption during detection and recovery. Existing fault-tolerance solutions intend to mitigate the limitations without considering critical conditions such as fail-slow faults, the impact of faults at various infrastructure levels and the relationship between the detection and recovery stages. This paper analyses the response time under two main conditions: fail-stop and fail-slow, when they manifest with node, service, and the task at runtime. In addition, we focus on the relationship between the time for detecting and recovering faults. The experimental analysis is conducted on a real Hadoop cluster comprising MapReduce, YARN and HDFS frameworks. Our analysis shows that the recovery of a single fault leads to an average of 67.6% response time penalty. Even though the detection and recovery times are well-turned, data locality and resource availability must also be considered to obtain the optimum tolerance time and the lowest penalties.

Link prediction analysis becomes vital to acquire a deeper understanding of events underlying social networks interactions and connections especially in current evolving and large-scale social networks. Traditional link prediction approaches underperformed for most large-scale social networks in terms of its scalability and efficiency. Spark is a distributed open-source framework that facilitate scalable link prediction efficiency in large-scale social networks. The framework provides numerous tunable properties for users to manually configure the parameters for the applications. However, manual configurations open to performance issue when the applications start scaling tremendously, which is hard to set up and expose to human errors. This paper introduced a novel Self-Configured Framework (SCF) to provide an autonomous feature in Spark that predicts and sets the best configuration instantly before the application execution using XGBoost classifier. SCF is evaluated on the Twitter social network using three link prediction applications: Graph Clustering (GC), Overlapping Community Detection (OCD), and Redundant Graph Clustering (RGD) to assess the impact of shifting data sizes on different applications in Twitter. The result demonstrates a 40% reduction in prediction time as well as a balanced resource consumption that makes full use of resources, especially for limited number and size of clusters

Background: Natural disasters, pandemics, armed conflict, and economic upheaval inflict lasting harm on mental wellbeing, yet what determines who recovers and who deteriorates remains poorly understood across crisis types.Objectives: This review maps how crisis exposure shapes wellbeing across four crisis typologies, two surveillance modes (NLP-based and survey/administrative), and three impact directions (favorable, adverse, or mixed).Eligibility criteria: Peer-reviewed journal articles and conference papers published in English between 2016 and 2026, applying machine learning, NLP, or deep learning methods to a population exposed to one of the four crisis typologies, and reporting at least one measurable mental health outcome.Sources of evidence: PubMed/MEDLINE, Web of Science, Scopus, and IEEE Xplore, supplemented by forward and backward citation searching and targeted keyword searches.Charting methods: Factors were charted using the WHO Commission on Social Determinants of Health framework, adapted for the health system where the framework’s coverage is thinnest. They were sorted into tiers, subgroups, and subdimensions with maximum possible fidelity to the original framework.Results: Across 338 codeable determinants, Intermediary Determinants, particularly psychosocial and behavioral-biological factors, accounted for two-thirds of coded determinants, while NLP-based methods proved most effective at capturing structural and contextual signals (e.g., governance, discourse-based constructs) rather than the clinical burden measures that dominate closer to the individual. Favorable, protective constructs were both rarer and less computationally detected than adverse ones, concentrating almost entirely in mastery, self-concept, coping, and social capital. Forty-five factors were set aside as Outcome Measures because they described prior symptoms, diagnoses, or the study's own outcome construct, with 8 being circular with the study's stated target. Fewer studies examined economic downturns, and conflict studies had almost no NLP presence.Conclusions: These findings demonstrate this field has matured methodologically in adverse detection but remains structurally limited in reach. This is most evident in armed conflict contexts and in the surveillance of protective, resilience-oriented signals, highlighting the need for digital phenotyping, expanded NLP coverage, and wider adoption of explainable AI methods.