Logic Learning Machine creates explicit and stable rules stratifying neuroblastoma patients

BMC Bioinformatics

Table 4 Final classifier

Rule ID ^a		NB-hypo	INSS Stage	MYCN Status	Age at diagnosis (Years)		Predicted Outcome	Covering b (%)	Error ^C (%)	Fisher pvalue^d	Stabilty^e	Risk group ^f
4.1	IF (	_	{4}	_	≥1	) THEN	Poor	91	4	<0.001	1	HR
4.2	IF (	High	{2, 3, 4}	_	≥1	) THEN	Poor	86	3	<0.001	1	HR,IR,LR
4.3	IF (	Low	_	Normal	_	) THEN	Good	89	0	<0.001	1	LR,IR
4.4	IF (	_	{1 2 3 4s}	Normal	_	) THEN	Good	90	0	<0.001	1	LR IR

^a The rule ID is composed by the table number followed by a dot and the rule number.
^b Covering is the fraction of examples in the training set that verify the rule and belong to the target class.
^C Error is the fraction of examples in the training set that satisfy the rule and do not belong to the target class.
^d Fisher p-value quantifies the statistical significance of a rule.
^e Stability is the fraction of the occurrences of a given rule in a 5 rounds of 10 folds cross validations.
^f The risk assessment is based on INRG classification taking into consideration only INSS stage, MYCN status and Age at diagnosis. HR = High Risk; IR = Intermediate Risk; LR = Low Risk.

ISSN: 1471-2105