Solving clustering problems using bi-objective evolutionary optimisation and knee finding algorithms

This paper proposes the use of knee finding methods to solve cluster analysis problems from a multi-objective approach. The above proposal arises as a result of a bi-objective study of clustering problems where knee regions on the obtained Pareto-optimal fronts were observed. With increased noise in the data, these knee regions tend to get smoother but still comprise the preferred solution. Thus, being the knees what decision makers are interested in when analysing clustering problems, it makes sense to boost the search towards those regions by applying knee finding techniques.