RPCA: A Novel Preprocessing Method for PCA

Algorithm 1

Pseudocode of ReliefF [2].

ReliefF Algorithm

(1) Initialization: given 𝐷={(𝑥_𝑗,𝑦_𝑗)}^𝑁_𝑗=1,  𝑦 is the label of classes between 1…𝑐.  

  𝑐 number of class, set  𝑤_𝑖=0,  1≤𝑖≤𝑡, number of iteration 𝑇;

(2) for 𝑙=1 to 𝑇

  (3) Randomly select a pattern 𝑥_𝑟 from 𝐷 with class𝑦_𝑟;

  (4) Find 𝑘 nearest hits 𝐻_𝑗 from class 𝑦_𝑟

  (5) For each class 𝑦≠𝑦_𝑟

  (6) from class 𝑦 find 𝑘 nearest misses 𝑀_𝑗(𝑦)

  (7) For 𝑖=1 to 𝑡

  (8) compute: 𝑤_𝑖=𝑤_𝑖−∑^𝑘_𝑗=1|𝑥_𝑟𝑖−𝐻_𝑗𝑖|𝑇⋅𝑘+∑_{𝑦≠𝑦𝑟}𝑝(𝑦)1−𝑝(𝑦_𝑟)^𝑘_𝑗∑₌₁|𝑥_𝑟𝑖−𝑀_𝑗𝑖(𝑦)|𝑇⋅𝑘

  (9) end

(10) end