Öğrenme tabanlı gazete satıcısı problemi

The newsvendor model is one of the most popular analytical models in decision science and operations management. The standard newsvendor problem is a single period inventory management problem in which the newsvendor has to decide the optimal stocking quantity for a single product. With the consideration of overstocking and understocking costs, the optimal quantity can be found that minimizes the newsvendor's total expected cost. Because of its simple structure, the newsvendor model has been widely applied to analyze several issues in supply chain systems involving perishable and seasonal products since the mid‐1980s. The standard newsvendor problem assumes the knowledge of the demand distribution so that the optimal stocking quantity is given by the critical fractile. In practice, however, this distribution is unknown; yet, there usually exist enormous historical demand data and demand related data. Then, the complexity of optimal ordering decision not only comes from coping with the ambiguity of demand, but also from incorporating a vast range of demand related information available into the decision process in order to make an enhanced decision. This research considers a data-driven newsvendor problem for a single product and a single period, where historical data of aggregated demands and attributes that can be used to leverage the demand distribution are available. Distribution of demand is assumed to be unknown except for its support. By modelling probability of buying the product through binary logit model, the demand process is approximated by a binomial process, and the resulting newsvendor problem is solved as a linear programming problem. Within the framework of this research, the robust approach is adopted as a benchmark problem, along with the Sample Average Approximation (SAA) approach. The separate numerical experiments for normal, gamma, and lognormal distributions show similarities with the well-known SAA-based optimization method.