Clustering of Quantitative Survey Data based on Marking Patterns

Abstract

Clustering of quantitative survey data is done in-order to identify the divergent and dominant behaviors of the respondents. It is intended to explore the general tendencies of the respondent groups. Popular clustering methods working on value based similarity are inappropriate for survey data due to its distinct properties. Since marking patterns in survey data represents respondent’s behavior, hence separating the responses on the basis of marking patterns is an effective approach to identify the dominant behaviors. Thus, in this paper, we propose a specialized clustering method for quantitative survey data that combines the features of both, value based as well as pattern based approaches in order to obtain meaningful results. The proposed method does not require presetting of the clustering parameters while it makes use of group labels for selecting features and guiding the centroids at positions, which best describe divergent marking habits. We apply the proposed method over an educational survey dataset and compare its results with K-means clustering with respect to the benchmark stakeholder theory. Comparison results show that the proposed method is more appropriate for quantitative survey data.