TIMESERIES BASED INVENTORY OPTIMIZATION SYSTEM FOR DISTRIBUTORS USING DEEP LEARNING

Abstract

Inventory control aims to meet customer demands at a specified service level while minimizing costs. Due to market volatility, customer demand often fluctuates, and ignoring this uncertainty can lead to incorrect inventory estimations, resulting in either shortages or inefficiencies. Inventory managers must place batch orders so that items arrive before stock depletion, considering the lead time between ordering and delivery. To meet demand while optimizing inventory costs, firms must forecast future demands to manage ordering uncertainties. Historically, predicting such uncertainties with high accuracy was challenging. However, the availability of large volumes of historical data and big data analytics has made this task more manageable.Finally, safety stocks are estimated based on the forecasted demand distribution, optimizing the inventory system to achieve a cycle service level objective. Using a comprehensive dataset from DataCo Global, which encompasses various supply chain operations such as provisioning, production, sales, and distribution, this thesis focuses on optimizing inventory for product categories including clothing, sports, and electronics. We employ a range of time series models— ARIMA (Auto Regressive Integrated Moving Average), SARIMA (Seasonal Auto Regressive Integrated Moving Average), Holt-Winters, and Prophet—to predict future demand with precision. The thesis begins with meticulous data preparation and cleaning to ensure data integrity. Exploratory data analysis (EDA) follows, revealing crucial patterns and relationships within the data, such as the effect of shipping days on delivery timeliness and the correlation between product prices and order quantities. Each time series model using mean absolute percentage error (MAPE) to determine their accuracy. The Prophet model, in particular, shows strong predictive performance, making it a valuable tool for guiding inventory decisions. Beyond forecasting, the thesis develops an inventory management framework. This includes calculating safety stock and reorder points based on demand forecasts, helping distributors maintain optimal inventory levels and avoid both stockouts and overstock situations. Visual tools display safety stock and reorder points, offering clear, actionable insights. The work is intended for various stakeholders in the distribution chain, including retail managers, manufacturing planners, logistics managers, sales and marketing teams, and corporate executives. By providing precise demand forecasts and actionable inventory management insights, this system enhances decision-making, operational efficiency, and profitability. This thesis demonstrates how time series forecasting can significantly improve inventory management for distributors. Implementing this system can lead to substantial cost savings, better customer satisfaction, and a stronger competitive position.