Density-based clustering defines a cluster as a dense region of points separated from other regions by emptier areas. It detects arbitrary shapes and isolates noise, which K-means does not. Its parameters, neighborhood radius and minimum number of points, depend on the scale of the data.