AN ENSEMBLE MODELLING APPROACH FOR PREDICTION OF FOOD PRICE IN AN ONLINE FOOD DELIVERY APPLICATION

Abstract

Online food delivery (OFD) services are increasingly common with a large restaurant industry share. Predicting the price of food items on the menu can be an interesting problem for machine learning methods. Data was extracted from a prominent OFD application. All food items from the menus of restaurants listed in the application were extracted. Seven most commonly used machine learning models (random forest, extreme gradient boosted tree, decision tree, bagged MARS, K-nearest neighbours, linear regression and support vector machine) were used for prediction. Based on the initial performance of the models, the best-performed candidate within each model type was used in the ensemble model. Of the 5,20,263 food items included in the analysis, 3,42,281 (65.8%) were vegetarian items. The ensemble modelling retained 16 out of 86 candidate members. The predictions of the candidate members in the ensemble were blended using lasso regression with a penalty of 0.001. When compared to the candidate models, the ensemble model performed slightly better (R square = 0.503 and RMSE = 0.505) than the highest performing random forest model (R square = 0.492 and RMSE = 0.511). Machine learning models like random forest and gradient boosted trees can predict food prices in OFD applications.

Received: February 16, 2022
Accepted: May 4, 2022