Product Demand Forecasting

Abstract

Accurate demand forecasting plays a crucial role in helping businesses maintain optimal inventory levels and ensure products are available when needed. This study investigates three widely-used time series models—ARIMA, SARIMA, and LSTM—for predicting product demand. The ARIMA model, a classic statistical approach, is effective in capturing the temporal dependencies within time series data, making it useful for forecasting demand based on historical patterns. The SARIMA model builds upon ARIMA by adding seasonal components, making it more effective for datasets with regular seasonal variations. On the other hand, Long Short-Term Memory (LSTM), a type of recurrent neural network, is designed to handle complex, non-linear data and is especially adept at modeling long-term relationships and irregular demand trends. This study evaluates these models using real-world data, examining their performance in handling trends, seasonal fluctuations, and demand anomalies. The evaluation focuses on the accuracy, efficiency, and appropriateness of each model for various forecasting needs. The aim is to provide businesses with insights into which model best suits their demand forecasting requirements, aiding in better decision-making, resource allocation, and inventory management. This research aims to offer valuable guidance for companies seeking to improve their demand forecasting and supply chain operations.