Artificial intelligence methods for classification and prediction of potatoes harvested from fertilized soil based on a sensor array response

Diet is one of the biggest vital reasons affecting human life, but the market is abundant with agricultural products treated with pesticides and fertilizers, which puts human health at risk. In this paper, the objective is to discriminate and predict the nature of potatoes (treated traditionally or with chemical fertilizers), one of the most consumed vegetables. To that aim, Potato samples were prepared and their headspace was analyzed. The data was collected by a data acquisition card from a sensor array manufactured with five commercial metal oxide (MOX) gas sensors from potatoes treated with NPK fertilizers, and from those treated traditionally with manure from domestic sheep and donkeys. The responses of the sensors were used to assess the accuracy of the classification and the prediction. Firstly, the k-nearest neighbors (KNN) with cross-validation of 5 folds were used for the classification and the accuracy was 90 %. Secondly, the nonlinear autoregressive with exogenous input (NARX) network was used for the prediction, and a correlation rate of 0.99 with a negligible mean square error. From the case studied, our tool is able to identify and predict whether the nature of the potato is treated with NPK fertilizer or with manure from domestic sheep and donkeys with very good rates.