Reported totals and fitted curves closely overlap. MAPE is 0.573% for wave 1 and 0.143% for wave 2, measured on the same cumulative data used to fit the model. These are in-sample fit errors; they do not measure future prediction accuracy or daily-case accuracy. Published metrics ↗
A close fit to cumulative totals still misses daily peaks. Wave 1 peaks at 5,743 reported cases on 21 October 2020, versus 3,847 simulated cases on 22 October. Wave 2 peaks at 9,317 reported cases on 11 May 2021, versus 10,170 simulated cases on 10 May. The daily model curve is the change in cumulative model output; its first day is omitted because no preceding model day exists.
Four compartments.
One evolving system.
Can become infected
Not yet infectious
Can transmit infection
Recovered plus deaths
The SEIR equations move people from S → E → I → R. RK4 combines four slope estimates to advance each day. Reported cumulative confirmed cases are tracked separately; they are not a fifth epidemiological compartment.
| Source | Johns Hopkins CSSE / Nepal time series |
|---|---|
| Preparation | Forward-fill missing values; recovered + deaths form removed counts; daily cases are non-negative differences of confirmed totals. |
| Fitting | 10 beta segments per wave / least squares / daily RK4 steps |
| Fixed assumptions | Exposed-to-infectious rate 0.192/day; removal rate 0.0628/day; recorded portion 0.0473 |
Initial states follow the reference setup for wave 1 and reporting-adjusted estimates for wave 2. Exposed and unreported infections are estimates. Data quality, reporting assumptions, and initial states affect the curves. Method adapted from Adhikari et al. ↗; data from Johns Hopkins CSSE ↗.
Beta represents the transmission rate in this model. A piecewise fit allows it to change over a wave instead of forcing a constant rate. The changes are inferred from the case data and model assumptions; they do not prove which particular intervention caused a rise or fall. Fitted beta values ↗
Each scenario scales every fitted beta segment while keeping the initial state and other parameters fixed. A 25% beta reduction lowers final simulated confirmed totals by 35.57% in wave 1 and 44.49% in wave 2, relative to the fitted baseline. These are conditional model outcomes, not observed cases prevented. They illustrate why transmission assumptions matter when exploring epidemic dynamics.
Graphs were redrawn directly from the repository’s published CSV outputs. No new model fitting was performed for this portfolio. Source data and results ↗