Neighbor-distance based diversity assessment for multi-objective optimizations

Performance assessment of Pareto-optimal set is an important issue in comparing or designing multi-objective evolutionary algorithms. This paper studies the characteristics and shortcomings of the current diversity metrics and proposed a distance and neighborhood-based uniformity indicator by using an improved K nearest neighborhood algorithm. In particular, this metric can compare the uniformity of population with different shapes and objectives. With an agglomeration of tests, the metric can accurately give an evaluation.