Indoor air quality (IAQ) monitoring helps detect pollutant buildup and mitigate unhealthy indoor environments. This paper presents an applied systems-integration framework that combines multivariate LSTM-based forecasting, ANN-based classification, rule-based health-risk assessment, and decision support in a single IAQ pipeline. The main contribution is this end-to-end functional integration rather than state-of-the-art predictive accuracy. The framework processes PM2.5, CO2, VOCs, temperature, and humidity through preprocessing, feature extraction, pollutant forecasting, IAQ classification, health-risk assessment, and decision-making. Evaluated on a synthetic multivariate sensor dataset, the LSTM achieved 89.7% forecasting accuracy, a PM2.5 RMSE of 4.2, a CO2 MAE of 35.6, and an R2 of 0.912. The ANN classified IAQ with 89.2% accuracy, compared with 78.5% for Random Forest. The framework also remained robust under missing data, noise, drift, rapid change, and seasonal shift. Because all results rely on simulated data, validation with real-world IoT measurements is required before deployment.
Evacuation guidance systems provide routes during emergencies, but occupants do not always comply when social cues compete with guidance. Limited research has examined attention allocation and sustained compliance across multiple decision points. This study compared environment-embedded cues (CON), AR navigation (AR-NAV), and a socially embodied AR virtual safety manager (AR-VSM) in an immersive VR evacuation experiment under competing environmental and social cues. Both AR conditions were simulated through a virtual smartphone interface. These conditions showed lower attention to surrounding non-player characters (NPCs), while AR-VSM yielded the highest sustained compliance. NPC-oriented attention was strongly negatively associated with compliance. An exploratory mediation analysis suggested statistical associations among guidance condition, attention, and compliance without establishing a causal mechanism. These findings highlight the relevance of attention allocation and guidance design to sustained compliance under competing local cues.
Pedestrian movement in high-density public spaces is essential for safe and efficient urban mobility, yet the system-level effects of movement heterogeneity remain insufficiently understood. Here, we develop a data-driven framework combining empirical trajectory analysis with an adapted social force model to investigate how non-mainstream behaviors affect pedestrian flows. A segment-based spatial cell method consistently identifies these behavioral deviations across real-world contexts. Empirical results show that non-mainstream behaviors increase local speed variability, reduce local flow, and are associated with greater congestion levels. Simulations reveal synergistic effects of crowd density and non-mainstream behavior: increasing density intensifies compression, while higher non-mainstream ratios introduce additional friction and weaken collective order, jointly amplifying congestion and crowd danger. Geometric interventions reorganize heterogeneous flows, mitigate conflicts, and improve efficiency and safety. These findings highlight the role of movement heterogeneity in crowd instability and inform spatial interventions for safer and more resilient pedestrian built environments.
