Predictive Modeling of Typhoid Incidence in Nepal
Master’s Thesis · Defense Presentation
May 2026
Master’s Thesis · Defense Presentation
Predictive Modeling of Typhoid Incidence in Nepal
Under extreme climate-change scenarios using machine learning
Kritika Baral
Master of Science in Health Informatics
Department of Health Informatics · Kathmandu University · Dhulikhel, Nepal
ORCID 0009-0007-1517-1063 · kritka.baral@gmail.com
A question of consequence
Can climate-induced flood events, precipitation, temperature, and humidity predict typhoid incidence across Nepal’s seventy-seven districts well enough to make public-health action anticipatory rather than reactive?
The setting
Nepal records between eighty thousand and one hundred thousand clinically diagnosed typhoid cases annually, concentrated in the flood-prone Terai plains. Existing surveillance is reactive: cases are counted after presentation — leaving no lead time for water purification, chlorination, or hygiene campaigns.
The scale of the work
77 Administrative districts of Nepal
108 Months · January 2015 – December 2023
7,327 Valid district-month observations
The first national-scale, district-level, four-model machine-learning study of typhoid fever in Nepal, integrating four validated open-access national datasets within a single analytical framework.
Four data foundations
| Source | Variable | Provider |
|---|---|---|
| HMIS | Monthly outpatient typhoid cases | Ministry of Health and Population |
| CHIRPS | Precipitation (5 km resolution) | Climate Hazards Group |
| ERA5-Land | Temperature, humidity (9 km resolution) | ECMWF · Copernicus |
| DRR Portal | Flood event counts | NDRRMA |
Bikram Sambat to Gregorian calendar conversion; CBS three-digit district-code reconciliation; area-weighted zonal aggregation of climate grids to district polygons.
Feature engineering
Eleven predictors, organized into five functional categories. All lag and rolling features are computed within each district to prevent cross-district data leakage.
| Category | Features |
|---|---|
| Autoregressive | Prior month’s case count |
| Lagged climate | Precipitation · Temperature · Humidity · Flood events |
| Cumulative climate | Three-month rolling precipitation · temperature |
| Seasonal indicator | Monsoon flag (June–September) |
| Cyclical encoding | Sine and cosine of calendar month |
Target variable: \(\log(1 + \text{cases})\) to stabilize variance over the heavy right tail.
Four architectures, one ensemble
| Model | Architecture |
|---|---|
| Random Forest | 500 trees · max depth unbounded · min samples leaf 2 |
| Extreme Gradient Boosting | 400 estimators · depth 6 · L1 + L2 regularization |
| Multilayer Perceptron | Hidden layers (128, 64, 32) · ReLU · early stopping |
| Weighted Ensemble | SLSQP-optimized weights: XGBoost 0.45 · MLP 0.35 · RF 0.20 |
Strictly chronological train–test partition. Ensemble weights are optimized on a held-out validation slice of the training set — never on the test set.
What carries the signal
| Feature | Pearson correlation with \(\log(\text{cases})\) |
|---|---|
| Prior month’s case count | 0.731 |
| Lagged mean temperature | 0.599 |
| Three-month rolling temperature | 0.559 |
| Lagged precipitation | 0.313 |
| Monsoon indicator | 0.245 |
| Lagged humidity | 0.180 |
| Lagged flood events | 0.122 |
Autoregression dominates. Lagged temperature is the strongest climate driver — consistent with the multi-pathway influence of temperature on pathogen biology, vector dynamics, and food safety.
Model performance — held-out twelve months
| Model | RMSE | MAE | R² | MAPE |
|---|---|---|---|---|
| Random Forest | 131.43 | 72.10 | 0.8563 | 41.48 % |
| XGBoost | 129.07 | 69.50 | 0.8614 | 44.64 % |
| Multilayer Perceptron | 128.07 | 71.45 | 0.8635 | 43.50 % |
| Weighted Ensemble | 126.20 | 68.07 | 0.8675 | 41.01 % |
The Weighted Ensemble achieves the best performance on all four metrics simultaneously — strong evidence that ensemble combination genuinely improves prediction beyond any single architecture.
The headline result
0.8675 R² · held-out twelve months
126.20 RMSE · cases per district per month
68.07 Mean absolute error · cases
At or above the upper bound of comparable climate-health machine-learning studies internationally. The integrated open-access dataset and ensemble framework constitute a technically reproducible and operationally feasible early-warning tool.
Seasonal pattern — the August peak
August registers the peak median incidence of approximately 640 cases per district per month — roughly three times the dry-season median. The interquartile range captures extraordinary spatial heterogeneity across the seventy-seven districts.
Geographic concentration
| Province | Districts | Monthly average | Monsoon peak | WASH coverage |
|---|---|---|---|---|
| Koshi | 14 | ~520 | ~1,850 | ~72 % |
| Madhesh | 8 | ~780 | ~2,400 | ~55 % |
| Bagmati | 13 | ~610 | ~1,650 | ~80 % |
| Gandaki | 11 | ~310 | ~850 | ~76 % |
| Lumbini | 12 | ~650 | ~1,950 | ~62 % |
| Karnali | 10 | ~220 | ~520 | ~48 % |
| Sudurpashchim | 9 | ~280 | ~680 | ~56 % |
Madhesh Province carries the heaviest burden — the convergence of lowest WASH coverage, highest flood exposure, flat topography that maximizes inundation extent, and high population density.
Projections under climate change
| Scenario | Precipitation | Temperature | Predicted cases | Δ |
|---|---|---|---|---|
| Baseline · 2015 – 2023 | 127 mm | 14.3 °C | 375,000 | — |
| SSP2-4.5 · moderate · 2050 | 146 mm | 15.3 °C | 469,000 | +25 % |
| SSP5-8.5 · high · 2050 | 165 mm | 15.8 °C | 525,000 | +40 % |
Monte Carlo propagation with one thousand draws; climate perturbations from CMIP6 regional projections.
These are indicative scenario-conditional projections, not precise forecasts.
Six policy recommendations
- Operationalize a district-level typhoid early-warning system.
- Prioritize pre-monsoon water, sanitation, and hygiene investments in high-burden Terai districts.
- Expand typhoid conjugate vaccine coverage in endemic districts of Madhesh, eastern Koshi, and Lumbini Provinces.
- Strengthen HMIS data quality and expand blood-culture diagnostic capacity to all seventy-seven district hospitals.
- Integrate climate-health monitoring into Nepal’s National Adaptation Plan 2021 – 2050 implementation framework.
- Establish a standardized post-flood disease surveillance protocol.
Limitations and future directions
Clinical-diagnosis surveillance, ecological design, and monthly temporal resolution all bound the present conclusions. Productive next steps include blood-culture sentinel validation, multilevel models incorporating individual-level WASH and vaccination indicators, downscaled CMIP6 burden projections under SSP2-4.5 and SSP5-8.5, weekly-resolution forecasting, spatial-panel and graph-neural-network architectures for inter-district propagation, and a systematic LSTM comparison against the four models presented here.
Acknowledgements
This work was made possible by the supervision of the Department of Health Informatics at Kathmandu University, and by the open-data commitments of the Ministry of Health and Population, the National Disaster Risk Reduction and Management Authority, the Climate Hazards Group at the University of California, Santa Barbara, the European Centre for Medium-Range Weather Forecasts, and the Central Bureau of Statistics of Nepal.
Closing
The predictive ceiling is not set by the choice of model, but by the climate–autoregressive signal already present in the data Nepal already collects.
The infrastructure for climate-informed early warning of typhoid fever therefore exists today.
What remains is the institutional decision to deploy it.
Kritika Baral · Master’s Thesis · 2026