Handling of Imbalanced Data in Text Classification: Category-Based Term Weights - Natural Language Processing and Text Mining - page 185

Table 10.9. Details of T-test on MCV1 , where alpha = 0.001 and degree of freedom
=34

All vs. TFIDF, alpha =0.001 CBTW 1 vs. All, alpha =0.001
Test t -Value t -Critical Test t -Value t -Critical

CC 21.509 3.354 TFIDF 30.171 3.354

ChiS 19.634 CC 10.688

IG 25.879 ChiS 13.465

OddsR 8.343 IG 6.571

RF 17.038 OddsR 24.003

CBTW 1 30.171 RF 13.368

Mean 20.429 Mean 16.378

StdDev 7.536 StdDev 8.882

Table 10.10. Details of S-test on Reuters-21578, P ( Z> = k )= p -Value, where two
F 1 values are the same if their difference is not more than 0.01
All vs. TFIDF CBTW 1 vs. All

Test n K p -Value Test n K p -Value

CC 12 11 3.174E-03 TFIDF 12 11 3.174E-03

ChiS 12 11 3.174E-03 CC 12 8 1.938E-01

IG 11 10 5.859E-03 ChiS 11 7 2.744E-01

OddsR 12 8 1.938E-01 IG 12 7 3.872E-01

RF 12 9 7.300E-02 OddsR 12 10 1.929E-02

CBTW 1 12 11 3.174E-03 RF 13 10 4.614E-02

Table 10.11. Details of T-test on Reuters-21578 , where alpha = 0.001 and degree
of freedom = 34
All vs. TFIDF, alpha =0.001 CBTW 1 vs. All, alpha =0.001
Test

t -Value t -Critical Test t -Value t -Critical

CC 19.571 3.467 TFIDF 28.893 3.467

ChiS 25.164 CC 8.587

IG 24.352 ChiS 3.682

OddsR 1.692 IG 3.993

RF 11.338 OddsR 22.619

CBTW 1 28.893 RF 15.139

Mean 18.501 Mean 13.819

StdDev 10.214 StdDev 10.326

better overall performance can be achieved if CBTW 1 is chosen as the weighting

scheme.

10.7 Conclusion

Handling of imbalanced data sets in TC has become an emerging challenge. In this
chapter, we introduce a new weighting scheme which is generally formulated as