Abstract
Accurate forecasting of future demand is essential for decision-makers and institutions in order to utilize the sources effectively and gain competitive advantages. Machine learning algorithms play a significant role in this mission. In machine learning algorithms, tuning hyperparameters could dramatically enhance the performance of the algorithm. This paper presents a novel methodology for optimizing the hyperparameters of Extreme Gradient Boosting (XGBoost), a prominent machine learning algorithm, by leveraging Artificial Rabbits Optimization (ARO), a recent metaheuristic algorithm, in order to construct a robust and generalizable forecasting model. Additionally, the study conducts an experimental comparison of ARO with two widely utilized metaheuristic algorithms, Genetic Algorithm (GA) and Artificial Bee Colony (ABC), by optimizing the eight different hyperparameters of XGBoost. For this experiment, 68,949 samples were collected. Furthermore, variables that have a significant effect on sales were investigated to enhance the reliability of the model. Ten independent variables, comprising a mixture of internal and external features including display size, financial indicators, and weather conditions, were identified. The experimental findings showcased that the implemented ARO-XGBoost model surpassed other implemented models, including the XGBoost model, Genetic Algorithm (GA) optimized XGBoost, and Artificial Bee Colony (ABC) optimized XGBoost models, across various evaluation metrics such as mean absolute percentage error. In summary, the use of artificial rabbits optimization, a recent metaheuristic algorithm, yielded satisfactory results for hyperparameter optimization of XGBoost. Furthermore, our proposed forecasting model is comprehensive and holds potential for serving as a valuable model for future studies.