Avoiding the Inherent Limitations in Datasets Used for Measuring Aesthetics When Using a Machine Learning Approach

Table 2

Average results presented in Figure 5, identifying hyperparameters and input size for each model.

Size	Model	Pearson	SD	Hyperparameters
16	GLMNET	0.5320	0.0717	Alpha=0
16	GBM	0.5234	0.0738	Interaction.depth=4, n.trees=500
16	k-NN	0.4831	0.0744	k=12; distance=2
16	SVM	0.5389	0.0713	Cost = 16 Gamma=0.00984
32	GLMNET	0.5451	0.0709	Alpha=0
32	GBM	0.5266	0.0733	Interaction.depth=4, n.trees=500
32	k-NN	0.4851	0.0750	k=12; distance=2
32	SVM	0.5581	0.0732	Cost=0.397 Gamma=0.00984
64	GLMNET	0.5406	0.0669	Alpha=0
64	GBM	0.5474	0.0723	Interaction.depth=4, n.trees=500
64	k-NN	0.4898	0.0752	k=12; distance=2
64	SVM	0.5503	0.0691	Cost=2.52 Gamma=0.000244
128	GLMNET	0.5473	0.0745	Alpha=0
128	GBM	0.5425	0.0679	Interaction.depth=4, n.trees=500
128	k-NN	0.4926	0.0720	k=12; distance=2
128	SVM	0.5687	0.0676	Cost=0.397 Gamma=0.00155
256	GLMNET	0.5555	0.0719	Alpha=0,15
256	GBM	0.5479	0.0704	Interaction.depth=4, n.trees=500
256	k-NN	0.4776	0.0774	k=12; distance=2
256	SVM	0.5778	0.0671	Cost=2.52 Gamma=0.000244
512	GLMNET	0.5748	0.0701	Alpha=0,15
512	GBM	0.5482	0.0765	Interaction.depth=4, n.trees=500
512	k-NN	0.4845	0.0758	k=12; distance=2
512	SVM	0.5747	0.0683	Cost=2.52 Gamma=0.000244
1024	GLMNET	0.5644	0.0708	Alpha=0,15
1024	GBM	0.5473	0.0685	Interaction.depth=4, n.trees=500
1024	k-NN	0.4908	0.0777	k=12; distance=2
1024	SVM	0.5782	0.0670	Cost=0.397 Gamma=0.000244
2048	GLMNET	0.5602	0.0733	Alpha=0,15
2048	GBM	0.5465	0.0692	Interaction.depth=4, n.trees=500
2048	k-NN	0.4482	0.0815	k=12; distance=2
2048	SVM	0.5723	0.0685	Cost = 2.52 Gamma=0.000244
fulldataset	GLMNET	0.5590	0.0719	Alpha=0,15
fulldataset	GBM	0.5476	0.0690	Interaction.depth=4, n.trees=500
fulldataset	k-NN	0.4299	0.0825	k=12; distance=2
fulldataset	SVM	0.5554	0.0721	Cost=2.52 Gamma=0.000244