Integrating machine learning and spatial clustering for malaria case prediction in Brazil's Legal Amazon

Area of Science:

• Epidemiology
• Computational Biology
• Public Health

Background:

• Malaria poses a significant global health threat, especially in Brazil's Legal Amazon.
• Environmental and socioeconomic factors in this region facilitate malaria transmission.
• Existing control methods are insufficient, necessitating advanced predictive tools for public health interventions.

Purpose of the Study:

• To evaluate six computational models for forecasting weekly malaria cases in Brazil's Legal Amazon.
• To identify the most effective model for malaria case prediction in the region.
• To assess the impact of spatial clustering on predictive accuracy.

Main Methods:

• Six computational models were assessed: Long Short-Term Memory (LSTM), Gated Recurrent Units (GRU), Support Vector Regression (SVR), Random Forest (RF), eXtreme Gradient Boosting (XGBoost), and Autoregressive Integrated Moving Average (ARIMA).
• Models were used to forecast weekly malaria cases across multiple states in the Legal Amazon.
• K-means clustering was integrated to account for spatial heterogeneity.

Main Results:

• The Random Forest (RF) model demonstrated superior performance, achieving the lowest Root Mean Squared Error (RMSE) and Mean Absolute Error (MAE) in most evaluated areas.
• Specific results for cluster 02 in Acre showed an RMSE of 0.00203 and MAE of 0.00133 using the RF model.
• Integrating K-means clustering enhanced the predictive accuracy of the machine learning models.

Conclusions:

• A hybrid approach combining machine learning models (especially RF) with K-means clustering offers a powerful tool for malaria surveillance.
• This method improves the understanding of localized transmission dynamics and spatial heterogeneity.
• The findings support enhanced public health strategies and targeted malaria control in high-risk areas.