Mean Shift
Best for: Mode/peak discovery
How it works
$$\mu^{(t+1)}=\frac{\sum_i x_i\,g\!\left(\left\|\frac{x_i-\mu^{(t)}}{h}\right\|^2\right)}{\sum_i g\!\left(\left\|\frac{x_i-\mu^{(t)}}{h}\right\|^2\right)}$$A non-parametric, mode-seeking procedure that slides each point uphill on a kernel-density estimate toward the nearest local maximum (mode) of the density. Each iteration moves a point to the kernel-weighted mean of its neighbours, $\mu^{(t+1)}=\frac{\sum_i x_i\,g(\|(x_i-\mu^{(t)})/h\|^2)}{\sum_i g(\|(x_i-\mu^{(t)})/h\|^2)}$, where $g=-K'$ is the profile of a kernel $K$ (typically Gaussian) of bandwidth $h$. Points converging to the same mode form a cluster, so the number of clusters is determined automatically by the density modes of the data.
Common fields
Computer vision · spatial clustering