Forecasting Retail Sales Demand by using AutoRegressive Integrated Moving Average and Graph Neural Network for Supply Chain Optimization

From the past few years, Supply Chains Management (SCM) plays vital role in globalization and complexity of industrial development. However, existing Machine Learning (ML)-based Extreme Learning Machine (ELM) system for demand forecasting in SCM making it difficult to understand complex relationships of customer information in retail sales due to its black-box structure. To overcome this challenge, this research proposes AutoRegressive Integrated Moving Average with Graph Neural Network (ARIMA-GNN) model to provide accurate demand forecast by capturing complex relationships in SCM of retail store. Initially, data is collected from Black Friday dataset comprises of sales transaction captured at retail store, which includes customer past purchase data and demographics. Then, preprocessed with mean imputation to handle missing values, Duplicate Row Removal (DRR) for removing duplicate rows based on all columns. After that, cleaned data transformed into common scale with standard scaler method, making it easier for demand forecast. The proposed ARIMA-GNN model utilized to predict future demand and sales by understanding relationship between products and customers, then optimizes supply chain operations based on predictions. This approach ensuring that inventory levels are maintained to meet customer demand and satisfaction. From the results, the proposed ARIMA-GNN outperformed the existing ELM in terms of (0.63).