Survey Assessment for Decision Support Using Self-Organizing Maps Profile Characterization with an Odds and Cluster Heat Map: Application to Children's Perception of Urban School Environments

Abstract

The interpretation of opinion and satisfaction surveys based exclusively on statistical analysis often faces difficulties due to the nature of the information and the requirements of the available statistical methods. These difficulties include the concurrence of categorical information with answers based on Likert scales with only a few levels, or the distancing of the necessary heuristic approach of the decision support system (DSS). The artificial neural network used for data analysis, called Kohonen or self-organizing maps (SOM), although rarely used for survey analysis, has been applied in many fields, facilitating the graphical representation and the simple interpretation of high-dimensionality data. This clustering method, based on unsupervised learning, also allows obtaining profiles of respondents without the need to provide additional information for the creation of these clusters. In this work, we propose the identification of profiles using SOM for evaluating opinion surveys. Subsequently, non-parametric chi-square tests were first conducted to contrast whether answer was independent of each profile found, and in the case of statistical significance (p ≤ 0.05), the odds ratio was evaluated as an indicator of the effect size of such dependence. Finally, all results were displayed in an odds and cluster heat map so that they could be easily interpreted and used to make decisions regarding the survey results. The methodology was applied to the analysis of a survey based on forms administered to children (N = 459) about their perception of the urban environment close to their school, obtaining relevant results, facilitating results interpretation, and providing support to the decision-process.