Deep learning networks have been shown to be useful predictive models of sequential data, such as audio, speech, and text. In particular, stacked layers of long short-term memory (LSTM) networks have markedly improved natural language processing capabilities. Stacked layers of 1D convolutions have also been shown to work well. Unfortunately, research into applying deep learning to consumer transaction data has been limited in scope.
Drawing on recent research into applying deep learning to a wider range of sequential data types, Eric Greene compares different approaches to creating models that predict payment amounts, time, and recipient for recurring expenses such as rent, loans, utilities, and services, outlining the data requirements, feature modeling, and neural network architectures that work best, as well as common issues in training and deploying deep learning networks. You’ll learn how to develop predictive models leveraging deep learning and terabytes of transaction data using LSTM and 1D convolutions.
Eric Greene is a principal architect for Think Big Analytics, where he brings technical innovation to fruition by working with business and technical leaders across industries. Eric has recently focused on cognitive computing applied to different domains within financial services and has designed and developed systems for real-time fraud detection, collections customer segmentation and automated prescription, customer account balance forecasting, and internal operations anomaly detectors.
©2017, O'Reilly Media, Inc. • (800) 889-8969 or (707) 827-7019 • Monday-Friday 7:30am-5pm PT • All trademarks and registered trademarks appearing on oreilly.com are the property of their respective owners. • email@example.com