基于LSTM-随机森林组合模型的电商平台产品需求预测

Inventory Demand Forecasting for E-Commerce Platforms Based on Combined LSTM-Random Forest Modeling

摘要: 精准的产品需求预测是电商平台优化供应链管理、降低库存成本及提升用户体验的核心环节。然而，电商场景中的需求数据具有高波动性、非线性和受多因素(如促销、季节性变化、用户行为等)影响的复杂特征，传统预测方法难以有效应对。本研究提出一种基于长短期记忆网络(LSTM)和随机森林(Random Forest, RF)的组合预测模型，通过融合LSTM的时序特征提取能力与随机森林的非线性建模优势，显著提升了电商场景下的需求预测精度。通过整合某电商平台四类核心产品(电子产品、服装服饰、家居用品、美妆个护)的历史数据，结果表明，组合模型在促销期突发需求和季节性波动场景中表现尤为突出，为电商平台的智能补货、库存优化及动态定价策略提供了有力支持。

Abstract: Accurate product demand forecasting is the core link of e-commerce platforms to optimize supply chain management, reduce inventory costs, and improve user experience. However, demand data in e-commerce scenarios has high volatility, nonlinearity, and complex characteristics affected by multiple factors (such as promotions, seasonal changes, user behavior, etc.), making it difficult for traditional forecasting methods to effectively deal with them. This study proposes a combined prediction model based on long short-term memory network (LSTM) and random forest (RF), which significantly improves the accuracy of demand prediction in e-commerce scenarios by combining the temporal feature extraction ability of LSTM with the nonlinear modeling advantages of random forest. By integrating the historical data of the four core products (electronic products, clothing, household products, beauty and personal care) of an e-commerce platform, the results show that the combined model is particularly prominent in the sudden demand and seasonal fluctuation scenarios during the promotion period, providing strong support for the intelligent replenishment, inventory optimization and dynamic pricing strategy of the e-commerce platform.