A comparison between PLSR, SVMR and NARX network for the mint treatment day prediction based on multisensor system

The ability to distinguish between edible aromatic plants treated with insecticides holds the attention of researchers in view of the toxicity of insecticides in human health. The malathion has a distinctive smell it an insecticide widely used to protect mint crops. In the present paper, three regression and artificial intelligence (AI)-based methods such as partial least squares (PLS) regression, support vector machine (SVM) regression, and the nonlinear autoregressive with exogenous input (NARX) were investigated to predict the mint treatment day with malathion. The data used in this work are collected using a multi-sensor system designed based on commercial gas sensors. In this case, the nonlinear autoregressive with exogenous input (NARX) was found the most effective achieving a correlation coefficient (R) of 0.99 with a very minimal mean squared error (MSE) of about 1.10288e-14. Thanks to the right choice of the appropriate algorithm, the mint treatment day could be predicted with a simple multisensor gas array.