Analysis and Prediction of Factors Influencing Delivery Time for Food Delivery

This paper looks into the delivery time issue in takeaway delivery services. As market competition rises and the need for timely deliveries grows, companies must use better delivery strategies. But things like uncertain delivery times, wider delivery areas, and traffic make it hard to control delivery efficiency. To deal with these, this paper uses machine learning (ML) techniques. It combines automatic feature selection and Stacking ensemble learning to predict takeout order delivery times. We preprocessed 45,593 training samples and 11,399 test samples. The TPOT automated feature selection method found 21 key metrics. This did better than traditional manual feature selection. Then, we tested six models, such as Decision Tree (DT), Random Forest (RF), XGBoost (GBoost), and K-Nearest Neighbors (KNN). After optimizing the parameters, the CatBoost model had the best performance in R² value, Root Mean Square Error (RMSE), and Mean Absolute Error (MAE). So, it was chosen as the regressor for the Stacking ensemble model. We made and compared several Stacking ensemble learning models. We improved the base model combination and hyperparameter optimization. The experimental results show that, compared to traditional ML algorithms, the Stacking ensemble model with RF, GBoost, DT, KNN, and CatBoost did very well in all evaluation metrics. The R² value went up by an average of 0.0316. The RMSE went down by an average of 0.3342. The MAE went down by an average of 0.2392. This greatly improved the prediction performance. This research gives a more accurate and practical way to predict delivery times in similar cases. It also helps improve delivery efficiency and customer satisfaction.