Visualizing Portfolio Tradeoffs With Perceptually Accurate Self-Organizing Maps

Abstract

Multi-objective optimization and unsupervised machine learning were used to visualize relationships between portfolio composition and business performance tradeoffs. A multi­objective global optimization algorithm reduced the exponentially large space of possible portfolio outcomes to a set of Pareto optimal tradeoff frontiers for various combinations of decision and outcome constraints. The high-dimensional decision space was then reduced with an unsupervised neural network to generate Kohonen self-organizing maps (SOMs) for each tradeoff frontier. The resulting decision SOMs were rendered with a perceptually accurate color map and interactively linked with the tradeoff frontiers to enable visual data mining. The interactive visualization revealed relationships between planning decisions and business outcomes in the complex and high-dimensional decision space. The interactive SOM automatically classifies similar strategies together and shows how neighboring strategies relate to each other. The analysis demonstrated that the proximity of portfolios on the tradeoff frontier does not imply similarity of underlying portfolio decisions, and that the mapping between decisions and outcomes becomes more complex with stricter constraints.