multivariate sales forecasting using gated recurrent unit network model

Market forecasting is an integral part of supply chain management. machine learning models have turned a new page in predictive analysis and helped organizations achieve improved accuracy. this paper focuses on creating a gated recurrent unit (gru) model to predict sales for multiple stores as a multivariate time series. grus are a variation of recurrent neural networks (rnns) used to sequence modelling tasks. the dataset used to create the model contains the unit sales of 3,049 skus sold in 10 stores. the sales data from the 3049 skus were grouped into the 7 departments to use as input to the model. a vector autoregression (var) and lightgbm models were used to compare the gru model predictions. mean absolute error (mae) and mean absolute percentage error (mape) were used to compare the 2 models. the mean mape values for forecasts of the gru, var, and lightgbm models were 13.77%, 14.87%, and 14.24% respectively, while mae values were 68 units, 72 units, and 69 units respectively. the study reveals that the gru model provides more accuracy for multivariate sales forecasting due to its ability to learn hidden patterns automatically and handle time mechanisms such as trends and seasonality.