Reduction of Decision Rules for Project Explanation on Public Project Portfolio

Abstract

The selection of a public project portfolio among a set of good portfolios directly impacts public organizations in a decisive manner. This task depends on the criteria of the decision maker and is especially hard when the solution is evaluated on many objectives. This combinational optimization problem has been addressed by multicriteria optimization algorithms which do not generate a single solution, but instead they generate a set of good solutions in the Pareto frontier. In this paper we propose to aid the recommendation of public projects with the utilization of simplified decision rules to explain the construction of the recommended project portfolio. Due to most decision problems can be represented in decision tables, we can form a decision table — in portfolio selection problem — with rules defined by the features of the projects (condition attributes) and the decision of support (the decision attribute). To facilitate the recommendation process, we simplify the decision rules via a hybrid rough set-based method that combines a genetic algorithm with an exact method. Less decision rules help the decision maker to faster analyze why some projects are accepted or rejected in a particular portfolio and —by this mean— get more certainty about his/her decision. With the proposed approach, there is a significant reduction of the attributes of the decision rules with a high accuracy of classification, which was proven on a set of the UCI repository for machine learning test.