As an important part of both the digital economy and the real economy, online retailing has developed rapidly in recent years. However, the rapid development of online retailing has resulted in serious product return problems for online retailers. The resale of returned products has a great influence on the inventory management of online retailers. This paper studies the inventory optimization problem based on product return forecasting. First, a product return forecasting approach using transaction-level data is proposed, which can effectively predict return quantities for a given future period. Then, a multi-stage inventory optimization model that incorporates product return forecasting is constructed, and the optimal ordering strategy is analyzed using dynamic programming. Finally, the impact of product return forecasting on cost is studied using actual data from a fast fashion apparel company. The numerical results show that incorporating return forecasting can lead to an average cost reduction of 6.5% for the company. Moreover, the value of product return forecasting for inventory management increases as the unit holding cost and the resale proportion of returned products rise.