A Hybrid Feature Selection Method Based on Rough Conditional Mutual Information and Naive Bayesian Classifier

Algorithm 1

Wrapper algorithm.

Input: data set 𝐷, candidate feature set 𝐶

Output: an optimal feature set 𝐵

(1)  Acc_𝐶= Classperf (𝐷,𝐶)

(2) set 𝐵={}

(3) for all   𝑓_𝑖∈𝐶 do

(4)  Score = Classperf (𝐷,𝑓_𝑖)

(5)  append 𝑓_𝑖 to 𝐵

(6) end for

(7) sort 𝐵 in an ascending order according to Score value

(8) while   |𝐵|>1 do

(9)   for all   𝑓_𝑖∈𝐵 according to order do

(10)   Acc_𝑓𝑖 = Classperf (𝐷,𝐵−{𝑓_𝑖})

(11)   if   (Acc_𝑓𝑖−Acc_𝐶)/Acc_𝐹>𝛿₁   then

(12)    𝐵=𝐵−{𝑓_𝑖}, Acc_𝐶=Acc_𝑓𝑖

(13)    go to Step  8

(14)   end if

(15)   Select 𝑓_𝑖 with the maximum Acc_𝑓𝑖

(16)   end for

(17)   if   Acc_𝑓𝑖≥Acc_𝐶   then

(18)    𝐵=𝐵−{𝑓_𝑖}, Acc_𝐶=Acc_𝑓𝑖

(19)   go to Step  8

(20)   end if

(21)    if   (Acc_𝐶−Acc_𝑓𝑖)/Acc_𝐶≤𝛿₂   then

(22)    𝐵=𝐵−{𝑓_𝑖}, Acc_𝐶=Acc_𝑓𝑖

(23)   go to Step  8

(24)   end if

(25)  go to Step  27

(26) end while

(27) Return an optimal feature subset 𝐵