An integrated data-driven method using deep learning for a newsvendor problem with unobservable features

• An inventory problem where random demand depends on certain observable features in addition to an unobservable feature is studied. A estimation has addressed first solve optimization problem. We propose integrated and approach using a deep learning network. Numerical examples show that the performs better than other approaches environments. consider single-period with both directly impact distribution. With recent advances data collection analysis technologies, data-driven classical management problems have gained traction. Specially, machine methods are increasingly being into problems. Although been developed for newsvendor problem, they often from available optimizing system separate tasks be performed sequence. One of setbacks this phase, costly cheap mistakes receive equal attention and, optimizer blind confidence learner its estimates different regions To remedy this, we newsvendor’s strategy facing complex correlated additional information about state system. give algorithm based integrating optimization, neural networks hidden Markov models use numerical experiments efficiency our method. In empirical experiment, method outperforms best competitor benchmark by more 27%, average, terms cost. further analyses performance set experiments.