Artificial Intelligence Enabled Demand Forecasting for Sustainable Fashion Retail

Abstract

The intensive growth of fashion e-commerce sites, the necessity of proper demand forecasting has grown in order to facilitate sound inventory control and make right time decisions in the retail business. The fashion industry is especially difficult to predict because of changing customer tastes, the circumstantial product life cycle and marketing trends. This paper suggests a hybrid demand forecasting model sustainable fashion retail based on a retail sales data collected at Kaggle, that contains both transactional and customer demographic characteristics. The data is formatted and processed to get valuable information about the customer behavior and the sales trends. The proposed method incorporates both the use of Random Forest and XGBoost in order to best reflect both linear and nonlinear data relationships. Regression metrics used to assess the model performance include the R2, RMSE, MAE, and MAPE. The experiments show that the hybrid model provides a better performance of R2 of 0.90, RMSE of 2.81, MAE of 2.27 and MAPE of 13.72. By contrast the predictive accuracy and error values of baseline models such as Linear Regression, LSTM, and Support Vector Machine (SVM) indicate low predictive accuracy and high error values and therefore indicate the success of the proposed hybrid approach.