Evaluating the 2022 Philippine Senate Election voting patterns extracted using hyperspectral unmixing

Abstract

Using precinct-level election data for the 2022 Senate election, we extracted several voting patterns using two hyperspectral unmixing algorithms. Candidates with higher vote share tend to have stronger affinity to an archetype or voting pattern. Among the extracted archetypes, we found correspondence to administration-allied candidates, and the opposition. Using the extracted patterns as basis, and following unmixing, we reconstructed the precinct level data to determine the relative reconstruction error (RRE). The VCA algorithm had higher RRE compared to the R-CoNMF algorithm. The latter algorithm was also able to accurately reconstruct the order or ranking of all the candidates. Consistent with expectation, we found that all winning candidates had low RRE, significant vote counts or high median, and narrow distributions or low IQR.