Machine learning techniques in disease forecasting: a case study on rice blast prediction

BMC Bioinformatics

Table 3 Overall comparison of multiple regression (REG), backpropagation neural network (BPNN), generalized regression neural network (GRNN) and support vector machine (SVM) based prediction accuracy of rice blast severity measured as average correlation coefficient (r), coefficient of determination (r²) and percent mean absolute error (%MAE) of observed value for 'cross-location' and 'cross-year' models.

Model(s)	Multiple Regression (REG)			Artificial Neural Network (ANN)						Support Vector Machine (SVM)
				BPNN			GRNN
	r	r ²	%MAE	r	r ²	%MAE	r	r ²	%MAE	r	r ²	%MAE
Cross-location models
2000	0.57	0.33	67.01	0.61	0.39	51.68	0.67	0.47	40.20	0.73	0.55	33.74
2001	0.44	0.23	77.95	0.50	0.29	66.42	0.62	0.45	64.51	0.70	0.54	48.93
2002	0.50	0.29	97.48	0.64	0.45	94.04	0.66	0.49	83.87	0.69	0.58	57.09
2003	0.48	0.26	66.76	0.58	0.34	58.03	0.68	0.48	53.28	0.82	0.67	41.37
2004	0.43	0.22	78.51	0.47	0.28	60.38	0.65	0.44	49.43	0.78	0.62	39.44
Average	0.48	0.27	77.54	0.56	0.35	66.11	0.66	0.47	58.26	0.74	0.59	44.12
Cross-year models
Location-I	0.42	0.21	65.96	0.51	0.29	61.12	0.60	0.37	55.93	0.69	0.48	47.97
Location-II	0.40	0.20	73.29	0.41	0.23	59.66	0.59	0.40	54.21	0.70	0.51	40.98
Location-III	0.42	0.22	56.11	0.62	0.39	38.41	0.76	0.59	33.10	0.78	0.62	25.85
Location-IV	0.68	0.49	68.29	0.73	0.55	50.85	0.78	0.63	41.44	0.85	0.73	34.90
Location-V	0.59	0.37	63.46	0.73	0.54	51.15	0.76	0.58	46.79	0.84	0.72	33.62
Average	0.50	0.30	65.42	0.60	0.40	52.24	0.70	0.51	46.30	0.77	0.61	36.66

where,
L – I = Location-I viz. Palampur (1^st date of transplanting; 15 days prior to normal transplanting)
L – II = Location-II viz. Palampur (2^nd date of transplanting; normal time of transplanting)
L – III = Location-III viz. Palampur (3^rd date of transplanting; 15 days after the normal transplanting)
L – IV = Location-IV viz. Rice Research Station, Malan (CSK HPAU)
L – V = Location-V viz. Farmers' fields, Pharer.

ISSN: 1471-2105