Beltel Datanomics has developed a forecasting model for intermittent demand goods for Forest

2022: Developing a forecasting model for products with intermittent demand

Beltel Datanomics on January 24, 2023 announced the development of a forecasting model for products with intermittent demand.

Beltel Datanomics was tasked with testing the applicability of automatic forecasting for products with intermittent and irregular demand, which include pneumatic equipment and fasteners supplied by Forest.

During the project, a research analysis of sales data was carried out, which showed:

• strong variability in data caused by the presence of a large number of zero-demand days;
• absence of pronounced annual seasonality in sales;
• anomalies in data in the form of a sharp increase in sales on one day, which is explained by a successful marketing campaign.

All these points indicate a high randomness factor in sales for most distribution center/commodity pairs and the difficulty in applying classical approaches to predicting time series. Therefore, a model for forecasting demand for two months and two quarters ahead was developed and implemented, which takes into account not only the change in sales volume, but also works with the frequency of demand for this position, and also clears historical data from abnormally large sales.

Model validation was performed over 12 periods of two months. For each test period, model quality metrics (MAE, SMAE, RMSE, SRMSE, mean deficit and surplus) were calculated. Cross-validation on 12 test examples showed better quality metrics for the constructed Datanomics model than simple statistical models such as Previous Value Prediction (Naive) and Exponential Smoothing Model (SES), for example for the SMAE metric by 15% and for the average surplus by 20%.

The ultimate goal of the project is to create an order to replenish all RCs with certain items for two months/quarters ahead.

{{quote 'We strive to use advanced technologies and automation of processes in order to increase the efficiency of the enterprise, - said Alexandra Sinitsyna, project manager of FOREST, - This project was an experiment for us. We wanted to test the applicability of automatic forecasting for goods with our demand characteristics. The project was not smooth - efforts were required at the data upload stage, but we confirmed the hypothesis and received a qualitative forecast result in the context of RC/product and detailing the next development steps. I would also like to note the high level of expertise of Beltel Datanomics specialists. }}