An AI-based approach to the prediction of water points quality indicators for schistosomiasis prevention

Abstract

We investigate the simultaneous daily prediction of the pH and temperature of a water point using AI-based methods. These parameters are part of the physicochemical parameters of surface water favoring the reproduction of parasitic worms responsible for Schistosomiasis. Wavelet Artificial Neural Network (WANN), Long Short Term Memory (LSTM) and Support Vector Regression (SVR) are the AI-based methods employed to build models with fifteen months collected data. They are evaluated through two metrics: root mean square (RMSE) and mean absolute error (MAE). The results show that in overall three methods give acceptable RMSE which varies from 1.59 to 0.17. WANN model shows the best performance with a RMSE equals to 0.17 and a MAE equals to MAE 0.12 over LSTM and SVR ones in forecasting parameters values one day ahead based on their two previous days observations.