Using Proximity Graph Cut for Fast and Robust Instance-Based Classification in Large Datasets

Table 2

Time and accuracy comparison of a proposed graph-based method implemented with NSW and HNSW graphs. Graph parameters are given as follows: dataset name (number of items, HNSW connectivity parameter ).

Graph parameters	Time (ms), accuracy (%)						Speedup NSW vs. 1-NN	Speedup HNSW vs. 1-NN
	1-NN		NSW-PATH		HNSW-PATH
Road signs (10K, M = 8)	209	92.57%	25.2	92.57%	72.4	92.57%	8.3	2.89
Road signs (10K, M = 64)	1615	82.57%	151	82.57%	186	82.57%	10.7	8.68
SUSY (5M, M = 4)	78.16	71.74%	22.41	62.74%	15.01	69.02%	3.49	5.21
SUSY (5M, M = 8)	89.5	71.87%	49.43	69.19%	21.43	70.58%	1.81	4.18
SUSY (5M, M = 128)	8538	71.86%	7271	71.12%	57.66	70.1%	1.17	148.1
HIGGS (11M, M = 32)	596.08	64.09%	253.61	62.27%	218.11	57.14%	2.35	2.73

Bold values refer to the best achieved speedup.