Time-series forecasting of seasonal items sales using machine learning – A comparative analysis

There has been a growing interest in the field of neural networks for prediction in recent years. In this research, a public dataset including the sales history of a retail store is investigated to forecast the sales of furniture. To this aim, several forecasting models are applied. First, some classical time-series forecasting techniques such as Seasonal Autoregressive Integrated Moving Average (SARIMA) and Triple Exponential Smoothing are utilized. Then, more advanced methods such as Prophet, Long Short-Term Memory (LSTM), and Convolutional Neural Network (CNN) are applied. The performances of the models are compared using different accuracy measurement methods (e.g., Root Mean Squared Error (RMSE) and Mean Absolute Percentage Error (MAPE)). The results show the superiority of the Stacked LSTM method over the other methods. In addition, the results indicate the good performances of the Prophet and CNN models.