Integration of Spatiotemporal Data in The Development of AI-fEaL: Artificial Intelligence for Early Warning of Leptospirosis in Negeri Sembilan, Malaysia

Abstract:

Leptospirosis is a zoonotic disease with a broad geographical distribution, occurring in both rural and urban areas of tropical, subtropical and temperate regions. Environmental, socioeconomic, demographic and weather factors result in considerable geographical and temporal variation in infection risk that is important to account for in the development of early warning systems. The current study investigated the association between these factors and the occurrence of leptospirosis in the Seremban district in Malaysia over a 10 year period using a combination of data mining and machine learning. Exploratory Data Analysis of the weather data demonstrated a relationship between leptospirosis occurrence with rainfall at lag between 12 to 20 weeks and temperature at lag 16 weeks. Using these weather variables as input, a neural network-based predictive model shows the most optimized performance in terms of accuracy, sensitivity, and specificity of 84.00%, 86.44%, and 79.33% respectively, which is an increase of 31.26%, 35.43% and 9.55% over the no lag model. Additionally, spatial analysis was conducted by using Principle Component Analysis (PCA) technique to investigate the potential inclusion of soil and landuse components for enhancing the predictive modelling. PCA has shown that high clay content and cation exchange capacity in the topsoil, as well as residential, recreational, and solid waste management land uses can indicate higher occurrences of leptospirosis.