Unlocking Inventory Efficiency: Harnessing Machine Learning for Sales Surge Prediction

Abstract

Purpose: Sales forecasting plays a crucial role in inventory optimization for retail stores, especially during special events such as promotions, advertisements, holiday season, weather, social and economic situations etc. These events drive significant changes in customer buying patterns. Some of these events are captured in the current forecasting models as part of trend, seasonality, and cyclicality. But many times, unexpected local events such as extreme weather conditions, riots, and regional events such as marathons, concerts have a significant impact on sales surges which are usually not captured in the sales forecast. This leads to inventory being out of stock and store managers placing last-minute manual orders. By accurately predicting these sales pattern changes, retailers can make informed decisions, ensuring optimal inventory levels and maximizing profits.

Methodology: In this paper, a data-driven solution was discussed that leverages machine learning models to predict sales pattern changes and surges during local events in a specific geographical location. Web scraping can be used to gather data on local events from a range of online sources, including Google News, local news channels, and social media platforms. By extracting pertinent details about upcoming events, it is possible to compile a thorough database of local happenings.

Findings: Data Analysis of historical sales data mapped to its local events can provide insights on key department-categories where there is a surge in sales. Machine learning models can be used to analyze, experiment and train historical sales data in conjunction with the events data.

Unique contribution to theory, practice and policy: Machine learning models would be trained to capture the complex relationships between different local events and their impact on sales. Therefore the study recommends using the machine learning approaches specified to consider various factors, such as event type, location, duration, and historical sales patterns, our models can effectively predict sales fluctuations during specific events.