Artificial intelligence in prediction of ecotoxicity of a series of s-triazine compounds as potential pesticides

Abstract

The elevated levels of pesticides and their residues occur in the environment lately due to increased usage of different agrochemicals. These pesticide residuals enter the human body through water and food. Nowadays different statistics and artificial intelligence tools are employed in order to solve different problems in agricultural science. Artificial neural networks (ANNs) have shown up as a convenient tool in establishing the non-linear mathematical relationships. The ecotoxicity of studied s-triazine pesticides was expressed as acute algae toxicity (AAT) and modeled by the ANN approach. Prior to applying feed forward multilayer perceptron (MLP) neural network with Broyden-Fletcher-Goldfarb-Shanno (BFGS) learning algorithm. The ANN modeling resulted in two networks with the best statistical performance. An excellent correlation was obtained between experimentally observed data and acute algae toxicity predicted data with correlation coefficient higher than 0.9342. Additionally, global sensitivity analysis (GSA) was conducted in order to estimate the influence of all molecular descriptors in the input layer on the networks performance.