Accuracy and efficiency of estimating nutrient values in commercial food products using mathematical optimization

Software that estimates nutrient values in commercial food products using the mathematical optimization techniques of linear programming and quadratic programming was developed. Dietary fiber and linoleic acid values were estimated for 31 food products for which an analytical value for one or both of these nutrients was available. Estimations were performed by three different nutritionists using three different methods-an existing trial-and-error method, the linear programming method, and the quadratic programming method. Estimation accuracy was determined by comparing the estimated nutrient amounts to the known nutrient amounts. When estimation error was expressed in grams per 100 g of product, the trial-and-error method was observed to be less accurate than optimization methods at estimating dietary fiber (P = 0.03). No statistically significant difference in accuracy was found between methods for dietary fiber expressed as percent error, or for linoleic acid when expressed per 100 g of product or as percentage error. The time required to complete ingredient amount estimations using optimization methods was significantly less than the time required for the trial-and-error method (P < 0.0001). Mathematical optimization increases the efficiency of estimating nutrient values and appears to be comparable in accuracy to existing methods for estimating nutrient values, though better methods for expressing nutrient estimation errors are needed to obtain more definitive conclusions.