Use of Attribute Driven Incremental Discretization and Logic Learning Machine to build a prognostic classifier for neuroblastoma patients

Table 7 Performance comparison among the configurations in the weighted classification on the test set.

Configuration ^f	Accuracy^a	Recall^b	Precision^c	Specificity^d	Negative Predictive Value^e
Base (not weighted)	80%	90%	82%	57%	72%
W26_74 (balanced outcome)	65%	63%	82%	70%	47%
Wl_1000 (bias poor)	66%	60%	86%	78%	47%
W1000_1 (bias good)	78%	98%	77%	35%	89%

^a Accuracy is the fraction of correctly classified patients and overall classified patients.
^b Recall is the fraction of correctly classified good outcome patients and the overall predicted good outcome patients.
^c Precision is the fraction of correctly classified good outcome patients and the predicted good outcome patients.
^d Specificity is the fraction of correctly classified poor outcome patients and the overall poor outcome patients.
^e Negative predictive value is the fraction of correctly classified poor outcome patients and the overall predicted poor outcome patients.
^f Configuration indicates the specific weights assigned to the outcomes in the weighted classification.

ISSN: 1471-2105