www-ai.cs.tu-dortmund.de/de/LEHRE/FACHPROJEKT/SS12/paper/concept-drift/scholz_klinkenberg_2006a.pdf
drift.dvi
︸ ︷︷ ︸
batch 1
, z(2,1), ..., z(2,m) ︸ ︷︷ ︸
batch 2
, · · · , z(t,1), ..., z(t,m) ︸ ︷︷ ︸
batch t
, z(t+1,1), ..., z(t+1,m) ︸ ︷︷ ︸
batch t+1
z(i,j) denotes the j-th example of batch i. For each batch i [...] := P (y = +1)
P (y = −1) ·
n∏
i=1
Lift (hi) Di
(hi(x) → y = +1)
Lift (hi) Di
(hi(x) → y = −1) (5)
This allows to compute estimates of the conditional probabilities as
P (y = +1 | x) ≈ β(x)
1 + β(x) . (6) [...] ∈ {−1, +1} indicating its classification. Data arrives over time in batches. Without loss of generality these batches are assumed to be of equal size, each containing m exam- ples.
z(1,1), ..., z(1,m) ︸ …