PrACTiC: A Predictive Algorithm for Chemoradiotherapy-Induced Cytopenia in Glioblastoma Patients

Table 1

The true positive of each class and accuracy of thrombocytopenia, neutropenia, and lymphopenia predicted by different machine learning models.

Toxicity type	Misclassification cost	Model	TP-class 1	TP-class 0	Accuracy	AUC
Thrombocytopenia	10	RUS boosted	75	85	85.6	0.88
	11		92	73	74.8	0.87
	12		85	73	73.9	0.84
Lymphopenia	4	Naïve bayes	78	79	78.9	0.83
		RUS boosted	71	84	81.7	0.80
		Naïve bayes	75	79	78.2	0.83
	5	Linear discriminant	71	70	70.4	0.75
		Naïve bayes	75	76	76.1	0.83
	6	Boosted tree	71	92	88.7	0.86
		RUS boosted	71	81	79.6	0.81
Neutropenia	9	RUS boosted	89	88	88	0.94
	10	RUS boosted	89	89	89.3	0.92
	11	RUS boosted	89	88	88	0.96

Bold values show the highest accuracy of toxicity prediction.