Application of artificial neural network in water quality index prediction: a case study in Little Akaki River, Addis Ababa, Ethiopia

Abstract

The Little Akaki River is one of the most polluted Rivers in Ethiopia as reported on many studies. These studies, however, mainly used concentration measurement of certain constituents and compared them against local and international standards. The multitude of objectives and differences in selected constituents in the various reports require scientific knowledge to understand the River water quality status. This may hinder policy makers and the public from knowing the extent of the River’s pollution. Water quality index is a useful method to get a summary of water bodies’ pollution extent. To this end, Canadian Council of Ministers of the Environment-water quality index approach was used. Furthermore, modeling of the index was performed using trained and validated artificial neural network. Twelve water quality parameters from 27 sampling sites in the Dry season (January/February, 2017) and Wet season (October/November, 2015) were used for index determination. Results show that, all sampling sites except one site in the upstream were under poor water quality category. Afterwards, the neural network model was trained and validated, for 12 inputs and one output, using several combinations of hidden layers (2–20), number of neurons in the hidden layers (5, 10, 15, 20, 25), transfer, training and learning functions. The most optimal model architecture was obtained with eight hidden layers, 15 hidden neurons that resulted in R2 value of 0.93. This shows a good agreement between calculated and predicted index values. Hence, an artificial neural network can be successfully applied for modeling Little Akaki River’s water quality index.