Thesis Format
Monograph
Degree
Master of Engineering Science
Program
Electrical and Computer Engineering
Supervisor
Grolinger, Katarina
Abstract
Vast sales data in the retail sector present opportunities to optimize operations and increase revenue. Consequently, studies have leveraged these data to predict sales, detect anomalies, segment customers, and similar. Anomaly detection techniques have been proposed to identify deviations in sales patterns at both the product and product-store levels. However, these studies did not consider sales patterns across similar stores, which could greatly benefit the retailers. For instance, if a particular store outperforms similar stores in sales of a specific product, best practices could be established and shared throughout the retail network, ultimately increasing overall sales. Consequently, this thesis proposes an unsupervised collective-contextual anomaly detection framework for the identification of outperforming stores in massive sales data from a large retail network. The proposed approach addresses the gap by decoupling the problem into two components. Collective anomalies are detected by leveraging autoencoders to identify patterns that deviate from normal behavior. Then, contextual anomaly detection leverages the designed similarity metric, which operates at the product level, to identify similar stores and examine patterns among them. Experiments using a dataset from a major Canadian retailer demonstrate the success of the proposed approach, achieving 90\% overall precision.
Summary for Lay Audience
Day-to-day sales transactions generate massive data repositories that can be leveraged through various techniques to improve retail operations and increase sales revenue. Sales anomaly detection is one such method widely used to identify unusual situations that provide valuable business insights. Additionally, major retailers with many stores recognize the importance of detecting anomalies when certain stores show higher sales performance for a given product. For example, if a specific product in one store sells much better than in similar stores, the successful strategies from that store can be shared to boost sales across all stores. While previous works have conducted sales anomaly detection at store and store-product levels, they have not analyzed sales across similar stores when identifying such sales anomalies. To address this gap, this work proposes a two-step method to identify these anomalies within massive sales data.
The proposed method first identifies unusual sales patterns of store-products that deviate from the norm using a deep learning model. Then, a product-level store similarity metric is introduced, which is computed based on monthly sales performance. Finally, the framework compares the detected unusual patterns against the most similar stores, using a mathematical measure defined to calculate total sales deviation, to determine outperforming store-product combinations. Several experiments were conducted to evaluate the performance of the proposed solution using a real dataset from a major Canadian retailer. This includes evaluating performance at different stages, analyzing the effect of framework parameters, demonstrating the importance of key model components using actual scenarios, and evaluating a case study. The results were verified by the industrial partner demonstrating the effectiveness and reliability of the proposed solution.
Recommended Citation
Fonseka, Thanuri Tehara, "Improving Retail Sales through Unsupervised Collective-Contextual Anomaly Detection: A Deep Reconstruction Autoencoder for Network-Wide Sales Analysis" (2024). Electronic Thesis and Dissertation Repository. 10566.
https://ir.lib.uwo.ca/etd/10566