Behavioral prediction in evolutionary objects has been, is, and will be a long-term subject of study. The number of data available to be processed increases, and so is the new Machine Learning approaches that target to improve the accuracy in their prediction rates, which opens a new way of automated processing in most systems by adding learning capabilities in most systems. This scenario starts revealing what will happen in the next decades, with intelligent systems automated in most societal areas. This article focuses on automatic trading and proposes modifying a conservative and general trend-based technique that discards prediction errors and learns from previous results. The introduced model predicts by finding patterns in those values responding well to the trending analysis, so future values with similar patterns can be successful trading assets. The study focuses on and compares three different models: The naive one with no learning part, a modification with the K-Nearest Neighborhood (K-NN) approach, and a different modification with an Artificial Neural Network (ANN). The results of the study turn out to be large and conclusive: Adding learning algorithms to this conservative technique boosts the overall accuracy when predicting successful trades.