Table 7. Total DBSCAN runtime in milliseconds when different NN search algorithms are used.

Dataset	eps	NMI	Brute force	$k$-d tree	Balltree	SNN
Banknote	0.1	0.05326	1,914	463.9	431.9	30.20
	0.2	0.2198	1,739	454.5	434.3	48.67
	0.3	0.3372	1,968	452.0	438.8	50.61
	0.4	0.5510	1,759	457.5	442.4	51.21
	0.5	0.08732	1,752	477.2	449.8	53.01
Dermatology	5.0	0.5568	706.8	138.8	124.3	86.81
	5.1	0.5714	654.0	142.2	127.8	64.62
	5.2	0.5733	651.8	139.2	123.2	63.74
	5.3	0.5796	650.7	138.2	121.5	60.75
	5.4	0.4495	638.6	138.4	121.2	58.17
Ecoli	0.5	0.1251	506.9	116.0	104.9	7.674
	0.6	0.2820	491.9	116.0	105.2	8.105
	0.7	0.3609	496.0	116.5	107.2	9.263
	0.8	0.4374	500.9	116.7	105.1	10.97
	0.9	0.1563	499.8	116.3	105.0	11.39
Phoneme	8.5	0.5142	3,497	17,290	7,685	926.9
	8.6	0.5516	3,511	17,480	7,738	954.1
	8.7	0.5836	3,300	17,490	7,727	937.9
	8.8	0.6028	3,257	17,600	7,768	975.2
	8.9	0.5011	3,499	17,570	7,734	1,065
Wine	2.2	0.4191	73.37	64.02	56.70	5.753
	2.3	0.4764	64.65	63.84	56.29	5.703
	2.4	0.5271	66.91	63.26	55.74	5.612
	2.5	0.08443	67.29	63.11	56.34	6.106
	2.6	0.07886	67.12	63.45	56.25	6.094

Note:

The DBSCAN radius parameter is eps and the achieved normalized mutual information is NMI. Best runtimes are highlighted in bold.