Automating the Ingredients Ordering Process in Restaurants: A Machine Learning-Powered System Approach

Abstract

In this research, we aimed to automate the ingredients ordering process in restaurants. By labeling and analyzing a subset of dishes from an Indian restaurant's menu based in London, we were able to accurately predict the quantities of ingredients needed in a given time period. This has the potential to greatly improve the efficiency of the ingredients ordering process, reducing the need for manual tracking and ordering and potentially reducing waste. We designed a pipeline consisting of three machine learning components, a detector, a determinator, a promoter, and a reducer. The machine learning components predict the total ingredients needed for different time frames, and the detector and determinator prevent overstock and understock, respectively. The promoter suggests possible actions in the event of overstock, and the reducer produces a preliminary order. Linear programming techniques could be used to finalize the order if constraints are present. The machine learning components were evaluated using the R squared score, and all three had a very high accuracy. The overall performance of the pipeline was evaluated using three custom metrics, which showed excellent results. This demonstrates that it is possible to automate the ingredients ordering process, even for small, non-franchised restaurants. While our study focused on a specific restaurant, the algorithm may also be applicable to other food service organizations, such as supermarkets and fast food chains. However, further research is needed to fully evaluate its performance in a wider range of settings.