Dual Stochastic Programming For Data Mining Enhancement

Parameter estimation problems represent one of the most common and widely used data mining sources, and they are usually carried out involving frequentist strategies supported in optimization formulations. Nevertheless, the results obtained from such optimization problems could not necessarily represent the most convincing solution. The shape, tendencies, and physical meaning of the obtained states profiles can even be questioned regardless the efficiency of the solving algorithms used. The proposal of a novel optimization structure (weighted lease squares based problem) solved by a dual stochastic algorithm in order to capture more detailed output profiles in a parameter estimation problem, is the main contribution of this paper. First trials assayed with data obtained from a lab scale dairy wastewater treatment process showed the benefits of using the developed techniques when comparing against traditional parameter estimation approaches. The dual optimization structure and the stochastic solver algorithm proposed in this work reveal to be a promising strategy for data mining enhancement.