Platelets are vital blood products required for various clinical treatments. Because of its perishable nature with a short shelf life (5–7 day) and cannot be manufactured synthetically, platelet inventory management is especially challenging and important. Motivated by the optimization of platelet ordering policies and inventory management, this study aims to build up a retrospective simulation of the platelet inventory to provide an optimized ordering strategy based on the forecasting of the platelet demand at University Hospital Jena (UKJ). The predictions of daily platelet demand were done with a Recurrent Neural Network (RNN) with Long Short Term Memory (LSTM) and statistical models (Ridge, Lasso, and Elastic net Regression). The predicted platelet demands are used as crucial input for Blood Inventory Simulation Model (BISM) to optimize platelet ordering and inventory strategies to reduce platelet waste and shortage. Compared with the actual expired waste of 3.37%, with the use of the optimized ordering strategy through BISM, the expired waste can be reduced to 0.79% with RNN LSTM and 0.83% with Lasso. After considering public holiday’s influence (as a feature) for blood platelet consumption, the expired waste can be even reduced to 0.61% with RNN LSTM and 0.69% with Lasso. The expired waste can be further reduced to 0.53% with RNN LSTM and 0.55% with Lasso with considering the different shelf life impact of platelet in stock for ordering strategy.