Title: Agent-Based Simulation of Informal Care Networks for Dementia Patients Under Changing Migration and Labor Policies

Abstract:Workforce shortages in long-term care interact with family migration patterns in ways static forecasts miss. We build an agent-based model linking household composition, wage differentials, and care-hour demand for community-dwelling dementia patients. Scenarios restricting caregiver visas increase unmet care hours and accelerate institutional placement, while modest stipends for co-resident relatives reduce hospital readmissions in sensitivity runs.


Title: Explainable Gradient Boosting for Early Warning of Student Dropout Risk Using Learning Management System Clickstreams

Abstract:Black-box risk scores erode counselor trust when interventions must be explained to students and families. Gradient-boosted trees trained on anonymized LMS clickstreams, assignment lateness, and forum participation are paired with SHAP explanations that surface actionable drivers per learner. A semester-long pilot flags at-risk students two weeks earlier than GPA-only rules, and advisors report higher willingness to act when feature attributions match observed engagement patterns.


Title: Participatory Mapping of Urban Heat Islands Coupled with Informal Vendor Cooling Strategies in Dense Market Districts

Abstract:Satellite land-surface temperature layers understate midday heat exposure where metal roofs and asphalt dominate open-air markets. Community mappers equipped with low-cost sensors and phone GPS co-produce fine-scale heat maps over eight weeks, while interviews document shade structures, misting, and staggered hours used by vendors. Overlay analysis identifies pockets where municipal tree planting and temporary canopy permits would yield the largest reductions in physiological heat stress.


Title: Edge-Deployed YOLO Variants for Real-Time Detection of Packaging Defects on High-Speed Bottling Lines

Abstract:Manual inspection at 400 bottles per minute misses intermittent seal and label faults that trigger costly recalls. Compact YOLO variants quantized for ARM-based edge GPUs are trained on a multi-plant defect corpus covering crooked closures, missing labels, and fill-level outliers. Online trials maintain precision above 0.94 at line speed with under 25 ms latency, enabling reject actuators to fire within existing conveyor spacing without slowing production.


Title: Classroom Acoustic Retrofit Using Recycled Textile Panels and Effects on Speech Intelligibility Scores in Secondary Schools

Abstract:Hard-surfaced classrooms elevate reverberation times that impair listening for students with mild hearing loss. Low-cost panels fabricated from post-consumer textile waste are installed on rear and side walls of twelve secondary classrooms. Measured RT60 drops into recommended bands, and word recognition scores improve significantly in bilingual instruction settings, demonstrating a circular-economy retrofit path for districts without capital for full acoustic ceilings.


Title: Graphene Oxide Membranes for Selective Lithium Recovery from Brine Concentrates After Conventional Evaporation Stages

Abstract:Evaporation ponds leave lithium-rich liquors still mixed with magnesium and calcium that complicate precipitation. Cross-linked graphene oxide laminates with tuned interlayer spacing are tested as selective nanofiltration layers on synthetic and field brines. Rejection of divalent cations exceeds 90% while lithium permeance remains industrially relevant over 200-hour continuous runs, cutting chemical consumption in subsequent carbonate precipitation relative to solvent-extraction baselines.


Title: Discrete Choice Experiments on Willingness to Pay for Community Microgrids Among Informal Settlement Households

Abstract:Tariff design for shared solar-battery microgrids often ignores how reliability and payment flexibility trade off for cash-constrained users. A discrete choice experiment with 640 households estimates part-worth utilities for outage hours, prepaid versus postpaid billing, and cooperative ownership. Preference heterogeneity clusters by employment type; evening-shift workers value reliability more than ownership stakes, informing tariff menus that raise uptake without subsidies that crowd out cost recovery.


Title: Biochar-Amended Compost Tea as a Foliar Biostimulant for Drought-Stressed Tomato Under Deficit Irrigation Schedules

Abstract:Deficit irrigation saves water but often cuts marketable yield when stomatal closure persists through fruit set. Foliar sprays of compost tea enriched with finely ground rice-husk biochar are evaluated in a two-season split-plot trial under 70% and 50% of full evapotranspiration. Treated plants maintain higher leaf relative water content and total soluble solids, recovering 12–18% of yield lost under severe deficit without increasing total seasonal water use.


Title: Transformer Models for Early Detection of Sepsis Onset from Continuous Vital-Sign Streams in Resource-Limited Wards

Abstract:Sepsis alerts based on static thresholds generate excessive false alarms when staffing ratios are thin. We train a temporal transformer on minute-level heart rate, respiratory rate, and oxygen saturation from two district hospitals and compare against NEWS2 escalation rules. The model advances median detection by several hours while keeping nurse workload within agreed false-positive budgets, although calibration drifts when transfer protocols change mid-study.


Title: Explainable Gradient Boosting for Early Warning of Student Dropout Risk Using Learning Management System Clickstreams

Abstract:Black-box risk scores erode counselor trust when interventions must be explained to students and families. Gradient-boosted trees trained on anonymized LMS clickstreams, assignment lateness, and forum participation are paired with SHAP explanations that surface actionable drivers per learner. A semester-long pilot flags at-risk students two weeks earlier than GPA-only rules, and advisors report higher willingness to act when feature attributions match observed engagement patterns.