Table 7.

Accuracy of different algorithms across five spatial and non-spatial models with outcome variable $d_i^{\mathrm{month}}$. For comparison, naive prediction of the majority class would achieve accuracy of 0.646 for New York and 0.702 for Los Angeles. The respectively best performing learning algorithm (LM, RF, GB, GP) is highlighted in bold.

	New York					Los Angeles
	NS1	NS2	S1	S2	S3	NS1	NS2	S1	S2	S3
LM	0.678	0.688	0.686	0.687	0.685	0.727	0.738	0.738	0.731	0.729
RF	0.680	0.679	0.656	0.665	0.649	0.729	0.733	0.707	0.698	0.702
GB	0.681	0.678	0.654	0.667	0.655	0.729	0.732	0.696	0.722	0.698
GP	—	—	—	—	0.690	—	—	—	—	0.742