Determining senatorial voting archetypes through hyperspectral unmixing

Abstract

Dimensionality reduction and hyperspectral unmixing techniques are commonly applied to process and analyze hyperspectral satellite images. In this work, we apply hyperspectral unmixing on the partial results of the Philippine senatorial elections last 2016 to determine possible voting archetypes. HySime was used for dimensionality reduction and determining the number of archetypes. VCA and MVSA, and SUnSAL were used to estimate the mixing matrix M, and the abundance matrix S, respectively. From the six archetypes obtained, the first archetype best describes votes for the 12 winning candidates. The second and third best describe the candidates that gained the most media coverage. The fourth archetype could represent celebrity voting. The fifth archetype could represent voters that choose candidates who have already run locally.