Artificial Neural Network for Prediction of Pollution Load of Lead, Copper, and Cadmium in a Water Resource: A case Study of River Sosiani, Eldoret Municipality, Kenya

James Obiewa; David Kariuki; Damaris Wachira-Mbui

James Obiewa Department of Chemistry, University of Nairobi
David Kariuki Department of Chemistry, University of Nairobi
Damaris Wachira-Mbui Department of Chemistry, University of Nairobi

Keywords: Artificial neural network, Correlation Pollution load, Water resource

Abstract

This study aimed at predicting the pollution load of Lead, Copper, and Cadmium in river Sosiani using the Artificial Neural Network, based on parameters, Physico-chemical; turbidity, Electrical Conductivity, and Chemical Oxygen Demand, and Chemical; fluoride and phosphate. The Atomic Absorption Spectrophotometer, Ultra Violet-Visible Spectrophotometer, Ion Selective electrodes and Redox-titration methods were used for analysis in from six sample sites, S1 to S6. A total of 78 datasets from the experimental results were used and divided into three, training 60%, testing 20%, and holdout 20%. The model used the IBM SPSS statistics 20 software, and performances evaluated using Pearson’s correlation coefficient. The mean pollution loads from laboratory analysis were 0.615±0.293, 0.037±0.027, and 0.096±0.030 mg/L while those from ANN were 0.615±0.293, 0.032±0.023, and 0.073±0.033 mg/L for Pb, Cu, and Cd, respectively. The correlation coefficients between the ANN and the observed values for Pb, Cu, and Cd were 0.9999, 0.9910, and 0.9965, respectively. The ANN was able to predict the pollution load of Pb, Cu, and Cd in the river.