Journal of Networking Technology Abraham Antonio López Villarreal, Samuel González-López, Luis Arturo Medina Muñoz 8 2 2017 http://www.dline.info/jnt/fulltext/v8n2/jntv8n2_2.pdf The purpose of this paper is to define the best training algorithm and activation function during the training process of a two degree of freedom planar robot arm that can be used to approach SCARA applications. A feed-forward neural network with two hidden layers was trained with several training algorithms such as Levenberg-Marquardt, Bayesian Regularization and others, and the activation functions such as symmetric sigmoid, logarithmic sigmoid and linear transfer to compare the resulting error and look for optimal performance.