Effects of input/output parameters on artificial neural network model efficiency for breakthrough contaminant prediction
Abstract Groundwater quality assessment is characterized by pollution injection rates, pollution injection locations and duration of pollution injection for identifying spatial and temporal variation. In this study, the spatial variations are obtained by placing observation wells in the downstream zone. And, the temporal variation of contaminant concentration has been simulated during the study period. Generally, simulations are carried out using various numerical models, which are subject to the availability of all required input parameters and are necessary for the proper management of contaminated aquifers. In literature, artificial neural networks (ANNs) are prescribed in such situations as these modeling methods focuses on available input-output datasets, thus resolving the concern of obtaining all inputs that numerical simulator usually demands. Past researches have predicted groundwater breakthrough contaminants. But the effects of input-output variations need to be discussed. This study is to quantify the effects of a few input-output datasets in the performance of ANN models to simulate pollutant transport in groundwater systems. The combinations of input/output scenarios have rendered these ANN models sensitive to variations, thus affecting model efficiency. These outcomes can reliably be employed for contaminant estimation and provide a paradigm in data collection, which will help hydrogeologists develop more efficient prediction models.