Rapid urbanisation and mounting climate risk have sharpened the focus on how urban green infrastructure sustains livability in increasingly data-driven cities. Although prior studies often isolate either spatial or digital dimensions, integrated evidence on their combined influence remains scarce. Addressing this gap, the present study investigated whether four complementary qualities of greenery, including spatial accessibility and connectivity, perceived physical quality, digital service integration and environmental value perception, jointly predict resident satisfaction. A questionnaire based on validated scales was distributed to residents of Ghent, Belgium, ensuring balanced representation across gender and age groups. Partial least-squares structural equation modelling with bootstrapping verified measurement robustness and estimated structural paths. All indicators exceeded accepted reliability and validity thresholds, and discriminant validity was confirmed through the Fornell–Larcker criterion and cross-loadings. The structural model explained the variance in satisfaction, revealing that walkable access to proximate, well-maintained green spaces produced the strongest positive effect, followed by moderate yet significant contributions from sensor-enabled services and recognition of ecosystem benefits. Our results underscore that walkable, high-quality green spaces provide the indispensable platform on which sensor-enabled services and compelling ecological narratives can truly thrive. To test the robustness and policy relevance of this framework, future studies should replicate it across a range of cultural and climatic settings. Ultimately, aligning place-based equity with digital innovation and environmental storytelling offers a scalable blueprint for building healthier, more resilient cities worldwide.
Distracted driving is becoming a growing problem in many countries, causing a huge number of road accidents and leading to death and serious injuries among drivers and pedestrians. Developing a distraction detection system has become a necessity to alert drivers and help reduce the number of road accidents. Thus, this paper presents a driver monitoring system designed to detect (1) the driver’s facial expression to identify whether they are drowsy or not, (2) facial front and sides monitoring, including hand movement, to detect (a) whether the driver is distracted or not, (b) a context-based distraction to specify the cause of such distraction. Our system utilizes lightweight image processing methods, including CNN-based models using EfficientNetB0. The model is efficient in achieving better accuracy with low computational resources, which makes it easier to put this system into real vehicles using mobile devices. We trained our model using a real-life dataset that contains numerous driver images in different situations to enhance its accuracy. The results show that the model achieved an accuracy of up to 95
This study presents a new sensor-integrated pen system that uses spiral drawing analysis to distinguish between essential tremor (ET) and Parkinson’s disease (PD). For the purpose of capturing motion and pressure data in real time, the system combines a tablet-based interface with a wireless electronic pen that is embedded with an accelerometer and gyroscope. Accurate tremor characterization is made possible by this dynamic handwriting analysis. A convolutional neural network (CNN) and an ensemble based on CatBoost were trained for classification; the ensemble performed better in PD detection and achieved 71.15
Urban resilience has evolved from a focus on physical infrastructure recovery to encompassing socio-economic, environmental, and governance dimensions. This paper presents an enhanced network-based framework for analyzing socio-economic resilience in cities, inspired by the Urban System Abstraction Hierarchy (USAH) model but extending it beyond acute shocks to address chronic stressors and multi-hazard contexts. We integrate insights from complex systems theory, socio-ecological resilience, and governance models to construct a comprehensive dependency matrix spanning seven domains, 40 indicators, and 451 sub-indicators. Using Vancouver as a case study, we mapped directed dependencies between indicators, enabling the identification of critical nodes through betweenness centrality. This approach highlights resilience bottlenecks where disruptions could cascade across multiple domains. We also conducted coupling analysis between domains to uncover interdependencies that may amplify risks. The results underscore the central role of Public Finance and Institutional Collaboration in sustaining city-wide resilience, suggesting targeted interventions for policy and investment priorities. The findings provide actionable insights for policymakers, urban planners, and stakeholders, offering a scalable methodology for other cities. This framework can guide resilience retrofitting, cross-sector collaboration, and informed decision-making, ultimately fostering adaptive, inclusive, and sustainable urban systems.
The rising concern of career-related anxiety and depression among university students poses a serious mental health challenge. To address this, we propose an explainable AI framework designed to assist early detection of such mental health conditions. Our approach integrates two primary types of data: publicly available survey data from Kaggle and gesture-based data collected through video interviews of university students. These datasets capture diverse behavioral and emotional cues relevant to mental health assessments. We applied robust preprocessing techniques tailored to each data modality. For structured survey responses, standard normalization and missing value handling were employed, while facial expression data such as head pose, gaze direction, and muscle movement intensities were processed using facial action coding and emotion feature extraction. An intermediate fusion neural network with attention mechanisms was utilized to effectively combine the two data streams. To improve model robustness, label smoothing was incorporated during training. To preserve user privacy and enable collaboration across institutions, federated learning (FL) was used, allowing model training without sharing raw data. For interpretability, we adopted Integrated Gradients and SHAP to identify important features influencing predictions. On the Kaggle survey data, our model achieved an accuracy of 91.76
