Machine learning approaches to supporting the identification of photoreceptor-enriched genes based on expression data

Table 5 The effect of class distribution on the performance of the classifier.

Class distribution (PR-enriched : non-PR-enriched)	Ac (%)	PR enriched			non-PR enriched
		Pr (%)	Se (%)	Sp (%)	Pr (%)
1:1	91.2	99.5	82.8	99.6	85.2
2:1	89.8	95.7	85.8	94.9	83.6
3:1	89.1	94.3	88.9	89.6	80.7
4:1	84.1	88.8	91.2	58.3	64.6
Original (261:63)	81.2	86.2	91.2	39.7	52.1

The class distribution is obtained based on random over sampling method. A 10-fold cross validation was carried out to estimate the true classification error. Class distribution is represented as the number of PR-enriched tags against the number of non-PR-enriched tags. There are 261 PR-enriched and 63 non-PR-enriched tags in the original dataset. In all resampled datasets, the number of PR-enriched tags is as also equal to 261.

ISSN: 1471-2105