Using machine learning method to predict food waste in catering industry under high resolution: a case in Dongguan

Waste classification is comprehensively carried out in China as an important national-level policy, and the large amount and the wide range of food waste generation (FWG) cause problems in the collection, transportation, and treatment. This study has conducted the prediction of FWG from the catering industry under high resolution, and provided suggestions and insights for food waste management and treatment. Taking Dongguan as an example, a Back Propagation Network (BPN) model is used to predict FWG under different operation data, and based on the acquired theoretical FWG numerical distribution, the intervals used to divide FWG values are determined. Then a Random Forest (RF) model is applied to predict the FWG intervals of the restaurants in the Point of Interest (POI) data. FWG of 96,303 restaurants is predicted, and the predicted FWG from the catering industry is about 3,106 t per day. Variation of FWG in different categories of restaurants, the material flow of FWG at the restaurant level, patterns of FWG at the restaurant level, and spatial patterns of FWG at the city level are also investigated. Suggestions for improvement of food waste collection standard and source reduction of FWG, and insights into food waste collection and distributed treatment system are raised.

Graphical abstract