Performance of machine learning system to prediction of almond physical properties

The physical properties of almond kernel are necessary for the proper design of equipment for transporting, drying, processing, sorting, grading, and storage this crop. In this study, different models of ANNs with different activation functions were used to forecast surface area, volume, mass, and kernel density of almond. The results showed that multilayer perceptron network with tanh-tanh activation function as a goodness activation function can be estimated surface area, volume, mass, and kernel density with R2 value 0.983, 0.986, 0.981, and 0.982, respectively. Furthermore, the physical properties were fitted by regression relationships, the result showed linear regression method can be predicted surface area, volume, mass and kernel density with R2 value 0.979, 0.961, 0.945, and 0.791, respectively. Generally, the result showed neural network model had high ability to forecast the physical properties of almond than the linear regression method.