Estimation of groundwater heavy metal pollution indices via an amalgam of stack ensemble learning

The Densu Basin is in the southern part of Ghana, with varied sources of contaminants because of agricultural activities, rapid urbanization, and industrialization. This paper evaluates heavy metal levels and specifically develops a prediction model for decision making. The prediction model adopted is a combination of artificial neural networks (as a first model-learner) and gradient boosting regressor (as a meta-learner). The model performed well with a strong correlation of 99.0 % between the predicted and the actual pollution indices, with root-mean-squared-log-errors of 0.025 and 0.012 for heavy metal pollution index and heavy metal evaluation index, respectively. Also, the results show that Mn and Pb are the two main relevant features necessary for the prediction of heavy metal indices.