Parameter employed for different models.

Precise demand forecasting has become crucial for merchants due to the growing complexity of client behavior and market dynamics. This allows them to enhance inventory management, minimize instances of stock outs, and enhance overall operational efficiency. In Bangladesh, there is a significant lack of emphasis on demand forecasting to enhance corporate performance. In recognition of these difficulties, the study seeks to produce predictions by employing two statistical models and three machine learning models. The historical sales data was obtained from a restaurant in Bangladesh, and five specific products were chosen for the purpose of predicting sales. The models have been rated according to their average score of deviation from the optimal root mean squared error. The Multilayer Perceptron and Random Forest algorithms have attained the top two positions. Statistical models such as simple exponential smoothing and Croston’s method have exhibited superior performance compared to XGBOOST model. This study advances demand forecasting techniques in Bangladesh’s restaurant industry by providing valuable insights, comparing different approaches, and suggesting ways to improve forecast accuracy and operational efficiency, thereby demonstrating the practical relevance and applicability of the research to the reader.

随着消费者行为与市场动态愈发复杂，精准需求预测已成为商户的核心要务。此举可助力商户优化库存管理、减少缺货情况，并提升整体运营效率。在孟加拉国，当前对用于提升企业经营绩效的需求预测工作重视程度严重不足。鉴于上述困境，本研究拟通过两种统计模型与三种机器学习模型开展需求预测工作。本研究的历史销售数据取自孟加拉国一家餐厅，并选取五款特定产品开展销量预测。本次研究以各模型相对于最优均方根误差（root mean squared error, RMSE）的偏差平均得分作为模型评级依据。多层感知机（Multilayer Perceptron）与随机森林（Random Forest）算法位列前两名。诸如简单指数平滑法与克罗斯顿法（Croston’s method）等统计模型的表现优于XGBOOST模型。本研究通过提供极具价值的洞见、对比不同预测方法、提出提升预测精度与运营效率的可行路径，为孟加拉国餐饮行业的需求预测技术发展提供助力，同时向读者展现了本研究的实际应用价值与适用性。