The Prediction of Diatom Abundance by Comparison of Various Machine Learning Methods

Table 4

Performance comparison for various models using 10-fold cross validation.

			True Value		Accuracy	Sensitivity	Specificity	AUC
			Absence	Presence
Without Boruta algorithm	LOGISTIC	Absence	57	30	64.64%	65.52%	63.83%	72.19%
		Presence	34	60
	kNN	Absence	61	33	65.19%	67.00%	63.33%	74.66%
		Presence	30	57
	CART	Absence	58	33	63.54%	63.89%	63.33%	68.84%
		Presence	33	57
	ANN	Absence	65	29	69.61%	71.33%	67.78%	76.65%
		Presence	26	61
	RF	Absence	67	28	71.27%	74.00%	68.89%	76.28%
		Presence	24	62
	BAG	Absence	58	26	67.40%	64.11%	71.11%	74.44%
		Presence	33	64
	BOOST	Absence	54	32	63.54%	60.67%	66.67%	73.40%
		Presence	37	58
	SVM	Absence	61	36	63.54%	67.22%	60.00%	75.41%
		Presence	30	54
With Boruta algorithm	LOGISTIC	Absence	59	29	66.29%	67.05%	65.59%	73.61%
		Presence	32	61
	kNN	Absence	56	32	62.98%	61.56%	64.44%	76.34%
		Presence	35	58
	CART	Absence	60	37	62.43%	66.11%	58.89%	68.68%
		Presence	31	53
	ANN	Absence	66	32	68.51%	72.78%	64.44%	77.93%
		Presence	25	58
	RF	Absence	67	28	71.27%	73.89%	68.89%	76.01%
		Presence	24	62
	BAG	Absence	60	26	68.51%	66.22%	71.11%	75.77%
		Presence	31	64
	BOOST	Absence	56	28	65.19%	61.78%	68.89%	72.95%
		Presence	35	62
	SVM	Absence	60	36	62.98%	66.22%	60.00%	76.22%
		Presence	31	54