APPLICATION OF DEEP LEARNING IN PRODUCT DEMAND FORECASTING

• Main Author: Oktavio, Audric
• Format: Final Project
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/76309
• Institution: Institut Teknologi Bandung

With the growth of e-commerce there has been a change in people's buying behavior. People who previously had to buy products face-to-face can now more easily buy their needs online. When buying products online, people generally expect the items they ordered to be delivered quickly. The sellers of products must respond to these fast demands by ensuring they have sufficient stock in advance to meet future buyer demand. To provide stock effectively, sellers need to predict future demand for products. However, small and medium businesses often rely on approximating previous sales to make predictions. This can lead to inaccuracies and problems such as stock shortages or excessive inventory that exceeds people's purchasing power. This Final Project aims to construct a demand prediction model for general stores using advanced Deep Learning algorithms namely, Gated Recurrent Unit (GRU), Long Short Term Memory (LSTM), and Transformer. The primary objective is to revolutionize the method of demand prediction specifically for SMEs, ensuring precise and optimal stock management. Among the algorithms assessed, GRU exhibited the most promising outcomes, achieving a mean absolute percentage error (MAPE) of 19.94% when predicting demand based on data from existing general stores.