This study applies Seasonal Autoregressive Integrated Moving Average (SARIMA) models to predict malaria

incidence in Beitbridge district, Zimbabwe. Using monthly malaria case data from 2015-2022, we developed,

evaluated, and validated time series models to forecast malaria trends. The optimal SARIMA(2,1,1)(1,1,1)₁₂

model was identified through comprehensive diagnostic testing. Model validation showed strong predictive

performance with a Mean Absolute Percentage Error (MAPE) of 12.7% and Root Mean Square Error (RMSE)

of 18.6. Six-month forecasts revealed expected seasonal peaks in April-May with declining trends. These

findings demonstrate SARIMA modeling's utility for malaria surveillance and can inform targeted intervention

timing in resource-limited settings. This research provides evidence-based tools for enhancing malaria control

strategies in Beitbridge and similar endemic regions.

control over the past decade, the district continues to face resource constraints that limit intervention

capabilities.

Effective malaria control strategies require accurate prediction of disease incidence patterns to optimize

resource allocation and intervention timing. Time series forecasting methods, particularly Seasonal

Autoregressive Integrated Moving Average (SARIMA) models, have demonstrated considerable utility in

predicting infectious disease patterns in various settings (Zhang et al., 2019; Anwar et al., 2021). These

statistical models account for temporal dependencies, seasonal variations, and trend components inherent in

disease incidence data, making them well-suited for malaria forecasting in regions with pronounced seasonal

transmission.

Previous studies have employed SARIMA models for malaria prediction in several African countries,

border crossing point with South Africa, experiencing high volumes of population movement that may affect

malaria transmission patterns. Additionally, the district's variable climate conditions, characterized by distinct

wet and dry seasons, create cyclical patterns in vector breeding that influence disease incidence (Zimbabwe

Meteorological Services Department, 2021). Understanding and predicting these patterns is essential for

effective malaria control in this region.

This study aims to develop and validate a SARIMA model for predicting monthly malaria incidence in

arid, with average annual rainfall of 350-450mm occurring primarily between November and March.

Temperatures range from 14-25°C in winter (May- August) to 22-34°C in summer (September-April). The

district is traversed by the Limpopo River and contains several seasonal streams that provide breeding sites for

Anopheles mosquitoes, primarily Anopheles arabiensis and Anopheles funestus, the main malaria vectors in

the region (Zimbabwe National Malaria Strategic Plan, 2021-2025).

Monthly malaria case data for Beitbridge district from January 2015 to December 2022 (96 months) were

obtained from the Zimbabwe National Health Information System (NHIS) and the district health information

offices with appropriate permissions. The data represent laboratory- confirmed malaria cases reported through

the national surveillance system. Ethical approval for use of these anonymized aggregated data was granted by

data were then organized chronologically to create a continuous time series for analysis.

Temporal trends and seasonal patterns in the malaria case data were explored using time series plots, seasonal

subseries plots, and autocorrelation functions. The data were decomposed into trend, seasonal, and irregular

components using the seasonal decomposition of time series by LOESS (STL) method. Seasonal patterns were

further examined using month-wise box plots to identify peak transmission months.

The Box-Jenkins methodology was employed to develop the SARIMA model, which is denoted as

SARIMA(p,d,q)(P,D,Q)s, where:

The model building process involved three key steps:

After validation, the full dataset was used to fit the final model, which was then employed to generate six-

month forecasts (January-June 2023) with 80% and 95% prediction intervals.

The time series consisted of 96 monthly observations of confirmed malaria cases from January 2015 to

December 2022. The mean monthly incidence was 128.5 cases (SD = 102.7), with a minimum of 14 cases

(August 2018) and a maximum of 473 cases (April 2019). Figure 1 shows the monthly malaria case time

followed by March (mean = 218.7 cases) and May (mean = 187.5 cases). The lowest incidence was observed

during the dry winter months of July (mean = 45.6 cases) and August (mean =43.2 cases).

Figure 2: Boxplot of monthly malaria cases by month in Beitbridge district (2015-2022)

The time series decomposition (Figure 3) separated the data into trend, seasonal, and irregular components.

The trend component showed an overall slight increase in malaria cases from 2015 to 2019, followed by a

decline in 2020-2021 (coinciding with increased malaria control efforts and COVID-19 movement restrictions),

and a slight uptick in 2022. The seasonal component confirmed the consistent annual cycle with peaks in

March-May and troughs in July-August.

779.48

790.12

-385.74

778.95

793.47

-384.47

784.21

795.96

-388.10

Table 2: Parameter estimates for the SARIMA(2,1,1)(1,1,1)₁₂ model

= 0.506), confirming the absence of significant autocorrelation. The Shapiro- Wilk test indicated that the

residuals were approximately normally distributed (W = 0.978, p = 0.231). Figure 6 shows the histogram and

Table 3: Forecast accuracy metrics for the validation period

Zimbabwe's busiest border crossing with South Africa. The slight uptick in cases in 2022 may reflect the

relaxation of these movement restrictions.

Our six-month forecast for early 2023 suggests continued seasonal patterns with an expected peak in April, but

with potentially lower overall incidence compared to historical averages. This project aligns with the recent

Medical Director for Matabeleland South Province Dr Andrew F Muza. We also thank the technical staff at the