Infection status outcome, machine learning method and virus type interact to affect the optimised prediction of hepatitis virus immunoassay results from routine pathology laboratory assays in unbalanced data

Table 3 Specificity and sensitivity (%) of HBV and HCV immunoassay outcome prediction after decision tree ensemble analyses

(a) Measure	Raw	Scale	Log	Scale-log
HBSA specificity	53.91	54.46	54.41	54.41
HBSA sensitivity	62.22	59.82	59.82	59.82
HepC specificity	57.75	57.65	57.77	57.66
HepC sensitivity	63.19	63.45	63.08	63.31
(b) Measure	Raw	Scale	Log	Scale-log
HBSA specificity	68.57	68.82	68.80	68.57
HBSA sensitivity	46.83	46.91	46.83	46.83
HepC specificity	58.87	58.91	58.88	58.87
HepC sensitivity	63.40	63.34	63.34	63.37
(c) Measure	Raw	Scale	Log	Scale-log
HBSA specificity	54.45	54.59	45.74	45.74
HBSA sensitivity	61.43	61.43	70.20	70.20
HepC specificity	35.04	34.87	36.90	36.88
HepC sensitivity	80.37	80.84	76.53	76.53

Methods employed were (a) basic multiple, (b) majority multiple and (c) clear negative analyses (see Methods). Prior to accuracy analysis, explanatory variables were subject to one of four pre-processing methods: none (raw), scaling, logging and scale-logging. Scaling sets the range of each explanatory variable to a common range of 0 - 100. Logging uses natural logarithm transformation. Scale-logging uses a common range of 0 - 100 then takes the natural logarithm.

ISSN: 1471-2105