Sales Anomaly Detection Using Automatic Time Series Decomposition

Abstract

This study proposes an automated time series decomposition (ATSD) technique for sales anomaly detection using weekly ice cream sale data from Google Trends with the caption "(Mobile AL-Pensacola (Ft. Walton Beach) FL. In this study, an empirical approach based on automatic time series decomposition (ATSD) was used to detect anomalies in sales data. A historical quantitative weekly ice cream sales data between periods of 9 April 2017 to 3 April 2022 consisting of 261 data points was used. The process for discovering the anomalous sales point includes importing of the necessary libraries, data visualization, construction of the ATSD model, creating the time series components from original data (seasonal, trend, residual), calculating the estimated from the original, and finally, the extraction of the anomalous sales. Anomalies were first discovered during the data visualization stage, which revealed some abnormal sales in mid-2019, late-2019, and early 2020. However, after applying the ATSD, our model detected anomalies with the specific dates of these anomalies. Following the application of the ATSD, our results indicate that the proposed anomaly detection approach can reliably detect anomalies with the dates of these anomalies. Because this technique worked well with our data, we assume it will work with any time series data.