Cluster Analysis in Practice: Dealing with Outliers in Managerial Research

Análise de Clusters na Prática: Lidando com Outliers na Pesquisa Gerencial

ABSTRACT

Context:

in recent years, cluster analysis has stimulated researchers to explore new ways to understand data behavior. The computational ease of this method and its ability to generate consistent outputs, even in small datasets, explain that to some extent. However, researchers are often mistaken in holding that clustering is a terrain in which anything goes. The literature shows the opposite: they must be careful, especially regarding the effect of outliers on cluster formation.

Objective:

in this tutorial paper, we contribute to this discussion by presenting four clustering techniques and their respective advantages and disadvantages in the treatment of outliers.

Methods:

for that, we worked from a managerial dataset and analyzed it using k-means, PAM, DBSCAN, and FCM techniques.

Results:

our analyzes indicate that researchers have distinct clustering techniques for dealing with outliers accordingly.

Conclusion:

we concluded that researchers need to have a more diversified repertoire of clustering techniques. After all, this would give them two relevant empirical alternatives: choose the most appropriate technique for their research objectives or adopt a multi-method approach.

Keywords:
cluster analysis; outliers; k-means; DBSCAN; fuzzy clustering