A parametric \(k\)-means algorithm (Q2271693)

A technique of principal (self-consistent) point estimation is proposed. It is based on parametric estimation of a distribution and application of a \(k\)-means clustering algorithm to pseudo-data simulated from this estimated distribution. The asymptotic behaviour of this procedure is investigated. A finite mixture of normal distributions is considered as an example. Results of simulations and an application to the problem of determining optimal sizes and shapes of protection masks for men by Swiss army data are considered.