Algorithm 1 Pseudocode for the ReliefF method for optimizing sensor arrays

Input: for each training instance a vector of attribute values and the class value Output: the vector $W$ of estimations of the qualities of attributes Procedure: 1: set all weights $W\left(A\right)$ = 0.0; 2: for i =1 to $m$ do begin   randomly select an instance $R_i$   find $k$ nearest hits $H_j$;   for each class $C$ ≠ class($R_i$) do    from class $C$ find $k$ nearest misses $M_j\left(C\right)$; 3: for each featrue $A$ =1 to a do $$\begin{gathered} W(A)=W(A)-{\mathsf{\Sigma }}_{j=1}^k\mathrm{diff} (A,R_i,H_j)/(m\ast k)+ \\ \hspace{1em}{\mathsf{\Sigma }}_{C\notin \mathrm{class} (R_i)}\left[{\displaystyle \frac{p\left(C\right)}{1-p\left(\mathrm{class} (R_i)\right)}}{\mathsf{\Sigma }}_{j=1}^k\mathrm{diff} (A,R_i,M_j\left(C\right))\right]/(m\ast k); \end{gathered}$$ 4: end;