Optimization is static or dynamic, and unconstrained or constrained. Further, optimization is concerned with search for local or global optima. Static optimization as typically appearing in scoring, churn and risk is single stage and time independent. Either models or at least data for possible models are available. If models exists, either derivates exist or they don’t. If derivates exist, unconstrained optimization is typically first (e.g. gradient descent) or second order, and suitable for local optima. Under constrains, usually different search strategies must be adopted. If models don’t exist, optimization is usually curve or multi-dimensional model fitting, and quality of models depends on quality of data. Dynamic optimization is used for multistage and time-varying action and decision-making, as typically appearing in optimization problems related to consumption, investment, portfolio and pricing.

Optimization & Prediction project phases and steps:

  • Understand the nature of your optimization problem.
  • Organize your data and check availability of possible submodels.
  • Identify variables, states and parameters.
  • Evaluate alternatives for the objective function.
  • Formulate your optimization problem.
  • Create your equations.
OPTIMIZATION - Example Concepts and Techniques