Comparison of daily sales forecasting methods using deep learning and time series features

In modern economics and business, the importance of accurate time-series forecasting has increased. This study investigates daily sales forecasting methods to compare their performance in sales forecasting. We investigate the performance of five combinations of time series features and deep learning architectures. We used a large data set with millions of records from the beginning of 2013 to the mid-2017. The main proposed method is a multi-input deep learning model that works on time series and lag features. We used the metrics to evaluate the model's performance: MAE, and MSE. The experiment was applied, and the proposed method achieved good results.