This paper presents the results of using autoencoder-derived features, rather than hand-crafted features, for predicting rod pump well failures using Support Vector Machines (SVMs). Features derived from dynamometer card shapes are used as inputs to the SVM algorithm. Hand-crafted features can lose important information whereas autoencoder-derived abstract features are designed to minimize information loss. Autoencoders are a type of neural network with layers organized in an hourglass shape of contraction and subsequent expansion; such a network eventually learns how to compactly represent a data set as a set of new abstract features with minimal information loss. When applied to card shape data, we demonstrate that these automatically derived abstract features capture high-level card shape characteristics that are orthogonal to the hand-crafted features. In addition, we provide experimental results showing improved well failure prediction accuracy by replacing the hand-crafted features with more informative abstract features.