J. Filipe and J. Cordeiro (Eds.): ICEIS 2009, LNBIP 24, pp. 363–375, 2009. © Springer-Verlag Berlin Heidelberg 2009 A Service Composition Framework for Decision Making under Uncertainty Malak Al-Nory 1 , Alexander Brodsky 1,2 , and Hadon Nash 3 1 George Mason University, Virginia, U.S.A. 2 Adaptive Decisions, Inc., Maryland, U.S.A. 3 Google Inc., California, U.S.A. {malnory,brodsky}@gmu.edu, hadonn@gmail.com Abstract. Proposed and developed is a service composition framework for decision-making under uncertainty, which is applicable to stochastic optimiza- tion of supply chains. Also developed is a library of modeling components which include Scenario, Random Environment, and Stochastic Service. Service models are classes in the Java programming language extended with decision variables, assertions, and business objective constructs. The constructor of a stochastic service formulates a recourse stochastic program and finds the optimal instantiation of real values into the service initial and corrective decision variables leading to the optimal business objective. The optimization is not done by repeated simulation runs, but rather by automatic compilation of the simulation model in Java into a mathematical programming model in AMPL and solving it using an external solver. Keywords: Modelling for stochastic programming, Object-oriented simulation, Supply chain optimization, Decision Support Systems. 1 Introduction Decision support information systems and frameworks often employ simulation and/or optimization techniques to help decision makers to analyze complex problems and establish actionable recommendations. For example, simulation and optimization have been widely used to minimize costs or maximize profitability in diverse enterprises. Mathematical Programming (MP) and Linear programming (LP) in particular, have been commonly used to model wide range of supply chain optimization problems, such as production and inventory planning, blending products, and network routing. MP tools require constructing a mathematical programming model with decision variables, constraints, and an objective function, possibly using a modelling language such as AMPL [1] or GAMS [2]. Such models can be solved efficiently using existing MP solvers with well-established optimization algorithms, e.g., for Mixed Integer Linear Programming (MILP). Deterministic LP models require complete information and can not solve problems in situations where some data or parameters used in the objective function or the constraints is not available at the time of solving the problem. However, most of supply chain real-world problems, such as