Objectives: To construct regression models in order to predict the behavior of data produced by motion sensors and muscle activity sensors linked to a human arm, with the intention of coordinating the movement of a robot. Methods: We use a LSTM (Long Short-Term Memory) network setup as a transfer learning problem where a sequence vector is generated and then a probability distribution and a set of real values are output and trained with separate cost functions. Findings: Evaluations are based on the precision accuracy of the algorithms applied to the data provided by the Myo Gesture Control Armband (Thalmic Labs) sensor used by different people with equivalent arm movements. These evaluations show a predictable behavior of the movements independently of the user. Novelty: The use of deep learning to solve a difficult problem of predicting time series avoiding overfitting. Keywords: Data Models, Deep Learning, Myoelectric Signal, Predictive Models, Real-Time, Regression Analysis, Time Series