ML Wiki
Machine Learning Wiki - A collection of ML concepts, algorithms, and resources.

Decision Analysis

Decision Analysis

Decision Under Certainty

Let

  • $A$ be a finite set of alternatives (possible decisions)
  • $X$ be a set of consequences (usually some financial metrics)
  • $c: A \mapsto X$ a consequence function
    • $c(a) \in X$ is a consequence of implementing action $a \in A$

Problem:

  • to compare alternatives and find the optimal one
  • on the basis of their consequences

  • When $ A $ is very large - need Optimization techniques - When $x \in X$ is multi-dimensional, i.e. $x = (x_1, …, x_m)$ need to apply Multi-Objective Optimization and/or MCDA

For these models we make a strong assumption:

  • we can quantify the consequences of taking different actions with certainty

However this assumption is not always true

  • we often can face situations when consequences $c(a)$ of taking a decision $a$ are not known with certainty

There are two categories of decision analysis tools that help model this:

Decision Under Uncertainty

  • we are not able to asses the distribution, but we can list all possible scenarios

Methods

Decision Under Risk

  • $c(a)$ is not known with certainty, but we know the probability distribution on the set of $X$

Decision Trees

  • http://answers.mheducation.com/business/economics/business-economics/decisions-under-risk-and-uncertainty
  • http://ids355.wikispaces.com/Ch.+5s+Decision+Making - Questions and Answers

Sources