Abstract

Transductive inference has been introduced as a novel
paradigm towards building predictive classi¯cation models
from empirical data. Such models are routinely employed
to support decision making in, e.g., marketing, risk manage-
ment and manufacturing. To that end, the characteristics of
the new philosophy are reviewed and their implications for
typical decision problems are examined. The paper's objec-
tive is to explore the potential of transductive learning for
corporate planning. The analysis reveals two main factors
that govern the applicability of transduction in business set-
tings, decision scope and urgency. In a similar fashion, two
major drivers for its e®ectiveness are identi¯ed and empir-
ical experiments are undertaken to con¯rm their in°uence.
The results evidence that transductive classi¯ers are well
superior to their inductive counterparts if their speci¯c ap-
plication requirements are ful¯lled.

Download

Share

COinS