Privacy-Preserving Feature Selection with Fully Homomorphic Encryption

For the feature selection problem, we propose an efficient privacy-preserving algorithm. Let $D$, $F$, and $C$ be data, feature, class sets, respectively, where value $x(F_i)$ label $x(C)$ are given for each $x\in D$ $F_i \in F$. a triple $(D,F,C)$, problem is to find consistent minimal subset $F' \subseteq F$, `consistent' means that, any $x,y\in D$, $x(C)=y(C)$ if $x(F_i)=y(F_i)$ $F_i\in F'$, `minimal' that proper of $F'$ no longer consistent. On distributed datasets, consider as problem: Assume semi-honest parties $\textsf A$ B$ have their own personal $D_{\textsf A}$ B}$. The goal solve A}\cup D_{\textsf B}$ without revealing privacy. In this paper, secure algorithm based on fully homomorphic encryption, implement our show its effectiveness various practical data. proposed first one can directly simulate CWC (Combination Weakest Components) ciphertext, which best performers plaintext.