Predicting rice blast disease: machine learning versus process-based models

BMC Bioinformatics

Table 1 Comparison of performance of M5Rules and RNN models on different individual dataset combinations

Training data*	Test data	Train		Test
Training data*	Test data	r (a)	r (b)	r (a)	r (b)	MAE** (a)	MAE** (b)	r² (a)	r² (b)
k2016	k2015	0.91	0.84	0.62	0.81	0.55	0.54	0.38	0.66
k2016	s2016			0.78	0.67	0.41	0.85	0.61	0.44
k2016	p2015			0.37	0.43	0.62	0.25	0.13	0.18
k2015	k2016	0.95	0.92	0.68	0.60	0.53	0.68	0.46	0.36
k2015	s2016			0.64	0.70	0.52	0.61	0.40	0.49
k2015	p2015			0.31	0.29	0.69	0.82	0.10	0.08
s2016	k2016	0.98	0.87	0.40	0.57	0.65	0.59	0.16	0.32
s2016	k2015			0.38	0.66	0.61	0.59	0.14	0.44
s2016	p2015			0.24	0.21	0.68	0.75	0.06	0.04
p2015	k2016	0.80	0.75	0.67	0.65	0.57	0.63	0.45	0.42
p2015	k2015			0.44	0.49	0.46	0.38	0.19	0.24
p2015	s2016			0.35	0.45	0.76	0.80	0.12	0.20
Average values				0.49	0.54	0.59	0.62	0.27	0.32

*k Kalochori, s Seville, p Portugal; **Mean absolute error, (a) = M5Rules, (b) = RNN

ISSN: 1471-2105