Predicting Sales Trends using Temporal Convolutional Network
Abstract
Retailers face crucial replenishment decisions such as which products to restock and when. To inform replenishment decisions, many retailers predict product demand using a rolling yearly average of product sales. Current AI demand forecasting solutions improve upon industry standards by capturing seasonality and promotional changes. However, these solutions are not effective for products that sell infrequently. Current AI solutions are also laborious, expensive, and time-consuming to implement due to the vast feature engineering required. Deep learning solutions are capable of circumventing lengthy feature engineering processes. Hence, a temporal convolutional network (TCN) was built to perform feature extraction. The TCN model architecture was designed and adapted to capture both short-term and long-term seasonal trends in the sales data, while also trying to learn price elasticity via promotional information.
About the Speaker
Sasha Nanda is a Senior Data Scientist at Deloitte, Omnia AI, where she builds machine learning models that address complex business needs such as demand forecasting, marketing experiment design, and customer segmentation. Sasha obtained a bachelor's degree in physics and minor in computer science from Caltech, where she specialized in quantum computing. She was a Feynman Quantum Resident at NASA Ames, and a Quantum AI Summer Resident at Google X. She then earned her Master's in Applied Computing at the University of Toronto, specializing in Data Science. Her thesis was on using a temporal convolutional network for demand forecasting in retail. Sasha is particularly interested in applying deep learning architectures to time series problems.
Register at https://www.fields.utoronto.ca/cgi-bin/register?form_selection=mathemati...