Sale forecasting in e-commerce using machine learning methods

Abstract

In recent years, there has been an increasing interest in the field of E-commerce and the predictions for its sale forecast. Future prediction is the process of making number forecasts which mainly bases on the historical data and current situation. The complication in real world data, hence, gives challenge for this procedure. This paper focuses on minimizing the errors of forecasting the sales of E-commerce stores in Vietnam by using several methods. First, certain traditional time-series forecasting methods are used, including Autoregressive / Seasonal Autoregressive Integrated Moving Average (ARIMA / SARIMA) and Triple Exponential Smoothing – Holt’s Winter methods. Next, more sophisticated technique is used – the Artificial Neural Network - Long Short-Term Memory (LSTM), Bidirectional Long Short-Term Memory (BiLSTM). Different accuracy measuring techniques, such as Mean Absolute Percentage Error (MAPE) and Root Mean Squared Error (RMSE), are used to compare the models' results. The results reveal that the Holt’s Winter Addictive approach is superior to the other methods. In addition, the results indicate the good performances of both LSTM and Holt’s Winter Addictive model.