A Comparative Study of Traditional and Machine Learning Approaches for E-Commerce Sales Forecasting

An increasing number of users are relying on online shopping in recent years. E-commerce is a rising field that shows great potential in business. A precise prediction on e-commerce sales could greatly affect the business plan of a company, influencing the marketing strategy, supply chain management, production, and many other aspects. Thus, e-ommerce sales prediction has become an attention-grabbing topic in the current business world. This study compares the performance of traditional statistical approaches and recent machine learning algorithms on e-commerce sales forecasting based on two datasets with and without time series data obtained from Kaggle. By comparing the performance metrics, including MAE and RMSE, machine learning methods outperformed in datasets without time series data, and statistical models achieved a better result in datasets with time series data. Thus, a trend that traditional models like ARIMA are better at time series data and machine learning methods have an obvious advantage in non-time series data has been proposed. Both of the models have their own strengths in corresponding fields.