Table 1.

Decision tree classification model cross-validation results

Test data parameters				Clean data		Raw data
Water years	Data points	Data size	ROS data size	Accuracy	F1 score	Accuracy	F1 score
2006 & 2007	212	27.8%	62.3%	0.976	0.981	0.825	0.871
2007 & 2008	125	16.4%	36.8%	0.992	0.989	0.816	0.875
2008 & 2009	64	8.4%	40.6%	1.000	1.000	0.891	0.868
2009 & 2010	82	10.7%	40.2%	0.951	0.943	0.951	0.943
2010 & 2013	124	16.3%	26.6%	0.960	0.928	0.823	0.667
2013 & 2014	134	17.6%	45.5%	0.948	0.942	0.821	0.774
2014 & 2015	73	9.6%	80.8%	0.863	0.911	0.836	0.891
2015 & 2016	46	6.0%	91.3%	0.891	0.943	0.804	0.883
2016 & 2017	191	25.0%	88.5%	0.979	0.988	0.853	0.913
2017 & 2018	184	24.1%	90.2%	1.000	1.000	0.875	0.927
2018 & 2019	99	13.0%	80.8%	0.939	0.962	0.747	0.839

A model was built and tested on two consecutive water years; the total test data side in number of points and percent of data are detailed along with the percentage of TWI events that were classified as ROS. Accuracy (number of correct classifications) and F1 score (harmonic mean of the precision and recall) is provided as a measure of model improvement for clean versus raw data.