Feature binary encoded patterns

Difficulties arising in applications of feature selection methods to binary encoded patterns have been discussed by Niyashita et.al. C2133. 

A feature selection method - called attribute inclusion algorithm C3653, ("attribute" is equivalent to feature) - for binary encoded patterns was applied to chemical problems by Schechter and Jurs C118, 2603. 

Tanimoto measure . The adoption of bit-encoded information led to the use of association coefficients for comparison of equal-length bit-vectors. In the case of our pattern-based similarity search, it is not the presence or absence of a feature that is being compared, but its frequency of occurrence. Discussions of the different types of similarity measures available for numerical, rather than binary, data can be found in books by Willett and Everitt. From these we chose to test the frequency Tanimoto coefficient and the Minkowski metric ... 

Substructure keys encode molecular information in the form of binary arrays or bitmaps (see Substructure Searching). Each element (or bit) in the array can take the values true or false , and indicates the presence or absence of a specific structural feature or pattern in the target molecule. Substructure keys were originally designed for large-scale database searching, but have also proven effective in similarity applications. 

---