Extracting voting patterns across three Philippine senate elections using hyperspectral unmixing

Abstract

We extract possible voting patterns by applying hyperspectral unmixing to the 2013, 2016, and 2019 Philippine elections. Hyperspectral unmixing, when applied to satellite images, extracts recurring traits or patterns found in a scene. First, data reduction determines the number of recurring patterns. Second, unmixing estimates the spectral signature of the recurring patterns. Lastly, inverting fits the obtained spectral signatures with the hyperspectral data to estimate their corresponding weights. By comparing the obtained voting patterns, we found the following recurring archetypes: opposition, conservatives, celebrities, political history, media popularity, and cultural-linguistic affiliation. The dominant archetypes for each province in each year were also calculated using their weights. We found that candidates tend to dominate their home province.