Machine Learning Application for Predicting Optimal Production Policy Parameters of a Production Unit Subject to Periodic Demand

This dataset accompanies the article "Machine Learning Application for Predicting Optimal Production Policy Parameters of a Production Unit Subject to Periodic Demand" (Operational Research in Engineering Sciences: Theory and Applications). It contains two datasets of 1,000 configurations each, generated by the numerical solution of the Hamilton–Jacobi–Bellman equations of a failure-prone production unit facing periodic demand. Each configuration gives the system parameters (mean demand rate and amplitude, and, in the second dataset, the unit backlog cost and the repair rate), the optimal critical thresholds at ten instants of the demand cycle and the optimal cost. The deposit also includes the test-set predictions of the seven regression algorithms compared in the article (k-NN, random forest, XGBoost, LightGBM, support vector regression, RBF network and deep neural network) over five repetitions, the corresponding performance metrics, and the values reported in Tables 5, 6, 8, 9 and 10. The file README.txt describes the generation of the data and every column.