Abstract
When we classify large-dimensional or geometric data, we generally use the smaller dimensional feature vectors instead of the original data. The feature vectors are extracted from the original data to avoid influence of rotation or parallel translation of data. A feature extraction method using Geometric Algebra (GA) is recently focused as a useful method to consider the geometric properties of objects. In particular, three types of features, inner product, coordinates and outer product, are often used in pattern classification with GA and higher effectiveness is expected by using the features together than using each independently. This paper proposes a new clustering method by combining GA and principal component analysis (PCA). Moreover, we verify effectiveness through some numerical examples.
