Sales forecasting in the fashion retail industry using machine learning techniques

Beh, Chi Qian (2024) Sales forecasting in the fashion retail industry using machine learning techniques. Final Year Project, UTAR.

Preview

PDF Download (5Mb) | Preview

Abstract

This project focuses on developing a sales forecasting system for the fashion retail industry to solve some problem including feature selection, algorithm limitations, and data visibility. Traditional statistical methods like SARIMA and Holt-Winters, along with machine learning techniques like LSTM, Prophet, and XGBoost, are evaluated for their effectiveness in sales forecasting. Data preprocessing including differencing and transformation are applied to ensure data stationarity, while feature engineering enhances model performance. Both daily and monthly forecasts have been developed and performance metrics show that the daily forecasts are more accurate than the monthly forecasts. Out of these models, XGBoost shows the best result compared to other models with the lowest forecast error and closest alignment with actual sales data. It also has been chosen for further analysis and deployment in the forecasting system. The findings derived from this project are useful in understanding the practice of machine learning tools in the sales forecasting of the fashion retail business and the role of model selection and preprocessing of data in enhancing the forecast results. In order to help users make informed decisions, the chosen model will be implemented into a web application that lets them input dates and evaluate prediction results.

Actions (login required)