PEMODELAN DAN OPTIMISASI MASALAH DEAD STOCK PADA SISTEM INVENTORI GEMBIRA HOUSEWARE MENGGUNAKAN DISCRETE EVENT SYSTEM DAN MACHINE LEARNING

• Main Author: Fidya Fachreza Istiawan, Adri
• Format: Final Project
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/61011
• Institution: Institut Teknologi Bandung

Gembira Houseware is a department store which focuses on selling houseware in Gorontalo Province. Gembira Houseware has as many as 7.256 product types with a hundred thousand monthly average number of transactions. Products that are sent from suppliers often could not be contained in the warehouse with warehouse utility reaches 80%. As many as 30% of these inventories are dead stocks or obsolete inventories. The objective of this study is to model the inventory system dynamic and propose an inventory policy to increase forecast accuracy and decrease number of dead stocks in the warehouse. Discrete Event System (DES) is used to model the dynamic inventory system. Demand forecasting process as an input of the modeled inventory system is modelled using machine learning algorithm to fit Gembira big data scale. Three forecasting models are used in this research, such as ARIMA, XGBoost, and LSTM. These models are used to predict demand of 12 product groups. It is found that each product group has its own best forecast model. ARIMA model gives the best performance overall with average MAPE score 16,60%. Based on inventory policies simulation, it is found that the proposed inventory policy can decrease number of dead stocks by 100%, number of inventories by 17%, and number of lost sales by 98%.