RBFNN-based model for heavy metal prediction for different climatic and pollution conditions

Heavy metal toxicity is a matter of considerable concern for environmental researchers. A highly cause of heavy metal toxicity in the aquatic environments is considered a serious issue that required full attention to understand in order to solve it. Heavy metal accumulation is a vital parameter for studying the water quality. Therefore, there is a need to develop an accurate prediction model for heavy metal accumulation. Recently, the artificial neural networks have been examined for similar prediction applications and showed great potential to tackle and detect its nonlinearity behavior. In this paper, radial basis function neural network algorithm has been utilized to investigate and mimic the relationship of heavy metals with the climatic and pollution conditions in lake water bodies. Thus, the present study was implemented in different climatic conditions (tropical “Malaysia” and arid “Libya”) as well as polluted and non-polluted lakes. Weekly records of physiochemical parameters data (e.g., pH, EC, WT, DO, TDS, TSS, CL, NO₃, PO₄ and SO₄) and climatological parameters (e.g., air temperature, humidity and rainfall) were utilized as an input data for the modeling, whereas the heavy metal concentration was the output of the model. Three different scenarios for modeling the input architecture considering the climate, pollution or both have been investigated. In general, results obtained from all the scenarios are positively encouraging with high-performance accuracy. Furthermore, the results showed that an isolated model for each condition achieves a better prediction accuracy level rather than developing one general model for all conditions.