Forecasting malaria cases in Harare province using Time Series Models

Author

Muvhu, Radwick.

Title

Forecasting malaria cases in Harare province using Time Series Models

Abstract

Malaria remains a serious public health problem in Zimbabwe. Reliable forecasting of malaria is crucial for effective resource allocation, early warning systems, and implementing targeted intervention strategies. This research study aimed to explore the application of time-series models in forecasting malaria cases in Harare Province, Zimbabwe. The objectives were to establish the trend of malaria cases, predict the future trend using SARIMAX and deep learning LSTM models, and to determine the best-performing model in terms of mean absolute percentage error (MAPE). Historical weekly data on suspected and positive malaria cases from 2013 to 2023 were analyzed to develop reliable forecasting models. Two-time series models were developed, evaluated, and compared based on their predictive performance to identify the most appropriate model for accurate malaria case forecasting. The results showed an overall decreasing trend in malaria cases in Harare province. For suspected malaria case prediction, the SARIMAX (3,1,3)(0,0,0,13) model achieved a MAPE of 20%, while the LSTM model demonstrated exceptional performance with a MAPE of 0.099%. However, for positive malaria case forecasting, both the SARIMAX (6,0,6)(0,0,6,13) and LSTM models struggled, with the LSTM model having a MAPE of over 259 million percent. The findings suggest the LSTM model is superior for predicting suspected malaria cases, but faces challenges in accurately forecasting positive cases. The SARIMAX model provided reasonably accurate forecasts for both suspected and positive malaria cases. Further research is needed to improve positive case prediction capabilities, potentially by exploring alternative model architectures.

Date

June 2024

Publisher

BUSE

Keywords

Forecasting

Malaria Cases

Time Series Models

Supervisor

Miss. J. Pagan'a