DBSCAN Parameterization – Lecture Notes

Suggested parameter for \(\text{minPts}=4\) is \(\varepsilon \approx 2.5\) With an index, this takes about \(n\cdot \log n\) time, \(O(n^2)\) without. On large data, we can query only a subset to make this [...] noted. Choosing DBSCAN Parameters Choosing \(\text{minPts}\) : usually easier to choose heuristic: \(2\times\) dimensionality [ EKSX96 ] choose larger values for large and noisy data sets [ SSEK17 ] Choosing [...] usually be \(\ll 1\%\) to \(\lt{}30\%\) , depending on the application) Choosing DBSCAN parameters /2 Example: \(\text{minPts}=4\) , therefore plot \(3\) -nearest-neighbor distances:     Sort the distances …