DengueOps AI
Simulation-Based Dengue Surge Preparedness Decision Support for Dhaka South
“Converting lag-aware dengue forecasts into supply depletion timelines, bed pressure estimates, uncertainty scenarios, and public health action priorities.”
Technical Positioning
This prototype does not claim a novel forecasting algorithm. Its contribution is the operational decision-support layer that translates outbreak forecasts into preparedness intelligence: supply depletion timelines, LOS-based bed pressure estimates, spatial zone priorities, and actionable directives.
The Problem
Dengue Response Is Often Reactive
In data-scarce urban health settings like Dhaka South, the gap between outbreak signals and operational response can cause preventable supply shortfalls and bed capacity crises.
Outbreak signals are recognised late
Surge pressure is often visible only after case load has already exceeded normal bed utilisation, leaving insufficient lead time for procurement or redeployment.
Supply shortfalls during peak weeks
Hospitals may exhaust NS1/RDT kits and IV fluids when demand spikes. Without stock-to-demand modelling, reorder decisions arrive after the critical window.
Fragmented surveillance, climate, and inventory data
Preparedness planning requires integrating dengue case counts, climate signals, facility capacity, and inventory — data that typically sits in separate systems.
Conventional dashboards stop at case prediction
Existing dengue prediction tools forecast case counts but do not translate forecasts into supply depletion timelines, bed pressure estimates, or facility-level directives.
Key gap: Forecast data is not being converted into supply or bed-load projections at the facility level. DengueOps AI is built to close this specific gap.
Solution
From Forecasting to Preparedness Intelligence
A Python analytics pipeline converts raw surveillance and climate data into operational readiness outputs — each step building on the last.
Dengue + Climate Data
Weekly epi case counts, rainfall, temperature, humidity
Lag-Aware Forecasting
14 & 28-day climate lags · governed Random Forest
Uncertainty Scenarios
Empirical forecast range plus separate planning sensitivities
Spatial Exposure Allocation
Zone priority score · vulnerability-gated heuristic
SDH + LOS Bed Pressure
Stock-to-demand horizon · length-of-stay bed load
Operational Directives
Per-facility: reorder, activate beds, vector control
Core Positioning
“DengueOps AI does not claim a novel forecasting algorithm. Its contribution is the operational decision-support layer that converts lag-aware outbreak forecasts into uncertainty-aware preparedness metrics and public health action priorities.”
System Architecture
Core System Modules
Six Python analytics modules, each producing structured JSON outputs consumed by the Next.js dashboard.
Lag-Aware Forecasting
analytics/forecast_model.py
Uses 14- and 28-day climate lags, case trends, and seasonality to generate a short-term synthetic demonstration forecast with the governed Random Forest selected in P1.2A.
Temporal Backtesting
analytics/validation_backtest.py
Uses chronological (time-based) train/test splitting, baseline comparison against naive and moving-average models, and reports MAE, RMSE, and MAPE to evaluate model behaviour.
Forecast Uncertainty
analytics/uncertainty_engine.py
Shows one prior-only synthetic empirical forecast range while preserving separate legacy planning scenarios. Not a prediction interval or probability guarantee.
Supply Depletion Horizon
analytics/operational_engine.py
Estimates how many days NS1/RDT kits and IV fluids may last under forecast-adjusted demand using a stock-to-demand horizon (SDH) model per facility.
LOS-Based Bed Pressure
analytics/operational_engine.py
Projects bed pressure using length-of-stay logic rather than treating beds as consumables. Outputs projected bed load and bed gap per facility under each scenario.
Operational Directives
analytics/operational_engine.py
Translates risk and readiness outputs into recommendations: reorder NS1/RDT kits, reorder IV fluids, activate additional beds, prepare referral protocol, prioritise vector-control response.
All modules produce synthetic demonstration outputs. Results illustrate pipeline behaviour, not deployment-grade epidemiological accuracy.
Differentiation
Not Just Another Dengue Prediction Dashboard
DengueOps AI extends beyond conventional case-prediction tools by translating forecasts into the operational metrics that preparedness decisions require.
Typical Dengue Dashboard
Case prediction · charts · risk level
- Predicts case counts
- Displays time-series charts
- Shows risk levels
- Stops at outbreak signal
- No supply depletion model
- No bed pressure estimate
- No zone priority framework
- No facility-level directives
DengueOps AI
Forecast → preparedness intelligence · decision-support
- Shows a temporally evaluated empirical forecast range
- Keeps Planning Low / Base / High scenarios separate
- Estimates NS1/RDT and IV fluid depletion horizon
- Projects LOS-based bed pressure per facility
- Ranks zones by vulnerability-gated exposure index
- Generates facility-level preparedness directives
- Provides model validation evidence (MAE, RMSE, MAPE)
- Transparent assumptions, human-review required
All outputs are advisory and require human review before any operational action. This is a simulation-based prototype, not a deployment-ready system.
Privacy & Ethics
Privacy-Safe Prototype Design
DengueOps AI is designed for use under public health data constraints. No patient data is required, collected, or processed.
No patient-level data used
All analytics run on aggregated epi-week case counts and area-level climate data. No individual records, identifiers, or clinical data are collected or processed.
Aggregated and synthetic data only
Where real surveillance data is not available, synthetic demonstration data is generated to illustrate pipeline behaviour. Data generation scripts are fully reproducible.
Real facility names as public-sector anchors only
Real public hospital names (DMCH, SSMC, Mugda General, NIBPS) are used only as spatial anchors where publicly available. General bed capacities reference public figures.
Dengue-specific values are synthetic
Dengue-specific bed allocation, current occupancy, NS1/RDT stock, IV fluid stock, and daily consumption values are synthetic demonstration values — not real clinical data.
Outputs are advisory — human review required
All directives, alerts, and readiness scores are decision-support aids. No autonomous operational actions are triggered. Human review is explicitly required.
Data Credibility Statement
Facility names and general bed-capacity anchors are based on public/government references where available. Dengue-specific bed allocation, current occupancy, NS1/RDT stock, IV fluid stock, and consumption values are synthetic demonstration values.
Evaluation Alignment
Designed for ICADHI Track 06
How DengueOps AI maps to IEEE ICADHI Track 06: Health Data Analytics & Predictive Systems evaluation criteria.
Technical Quality
Lagged climate features, chronological train/test split, baseline comparison (naive, moving average), MAE/RMSE/MAPE metrics, uncertainty engine.
Originality
Forecast-to-preparedness operational layer: SDH supply depletion, LOS bed pressure, zone priority scoring, facility-level directives — not only case prediction.
Functionality
Fully working Next.js dashboard visualising forecast, model validation, uncertainty, supply depletion, bed pressure, zone priority, and operational directives.
Impact
Supports dengue surge preparedness planning for hospital administrators and public health teams in high-burden urban settings like Dhaka South.
Scalability
Modular Python analytics pipeline designed for city-level scale-up. Can expand to other cities, diseases, or real data feeds with minimal architectural changes.
Ethics
No patient-level data, explicit assumption disclosure, transparent model limitations, synthetic data labelling, and human-review requirement for all outputs.
Users & Roles
Who Is This For?
DengueOps AI is designed around specific operational user needs, not generic dashboard consumption.
Public Health Officials
Primary operational users- ·Zone priority ranking and exposure indices
- ·Vector-control deployment recommendations
- ·Area-level surge risk and growth factor
- ·Early warning based on 14-day forecast
Hospital Administrators
Facility readiness users- ·NS1/RDT kit and IV fluid depletion timeline
- ·Projected bed load and bed gap estimates
- ·Facility-level priority and action directives
- ·Supply reorder threshold alerts
MIS / Data Officers
Technical pipeline operators- ·Run analytics pipeline when data is updated
- ·Validate CSV inputs against data contracts
- ·Monitor pipeline run logs and step outputs
- ·Maintain facility and inventory configuration
Technical Evaluators
IEEE judges · Researchers- ·Review model validation metrics (MAE, RMSE, MAPE)
- ·Inspect baseline comparisons and chronological split
- ·Examine uncertainty methodology and limitations
- ·Assess pipeline architecture and assumption transparency
Public / Citizens
Future target audience- ·Not the primary current user group
- ·A simplified public advisory layer is planned
- ·Current prototype targets institutional users
- ·Risk communication outputs may expand in future
Operational design principle: Operational users are not expected to code, clean CSV files, or run scripts during an outbreak. The analytics pipeline is maintained by technical/MIS staff, while hospital and public health users receive translated action recommendations through the dashboard.
Live Prototype Preview
Current Pipeline Output Snapshot
Metrics below are read from the latest analytics pipeline run. Values use synthetic demonstration data.
Forecasted Cases
120
Expected · 14-day horizon
Risk Level
High
Risk score: 71 / 100
Highest Priority Zone
Benchmark Zone 5
Vulnerability-gated exposure score
Facilities Monitored
10
0 real public anchors
Critical Supply Alerts
0
Items below 7-day threshold
Facilities — Bed Gap
0 / 10
Expected bed deficit (expected scenario)
Synthetic demonstration data · Advisory outputs · Human review required
Open Full DashboardIEEE ICADHI Project Showcase
Explore the Preparedness Dashboard
View the complete prototype dashboard with forecast uncertainty, model validation, zone priorities, facility readiness, supply depletion timelines, and operational directives — all generated from the analytics pipeline.
Synthetic demonstration data · Advisory outputs only · No patient-level data · Human review required before any operational action.
Made By
Authors
IEEE ICADHI Project Showcase · Daffodil International University
Meherab Hossain Shafin
AuthorJannatul Tazri Aohona
AuthorAvatar images are placeholders. · DengueOps AI · IEEE ICADHI 2025