Better Huffman Coding via Genetic Algorithm

Cody Boisclair and Markus Wagner


We present an approach to compress arbitrary files using a Huffman-like 
prefix-free code generated through the use of a genetic algorithm, thus 
requiring no prior knowledge of substring frequencies in the original 
file. This approach also enables multiple-character substrings to be 
encoded. We demonstrate, through testing on various different formats of 
real-world data, that in some domains, there is some significant 
advantage to using this genetic approach over the traditional Huffman 
algorithm and other existing compression methods.

Key words: Genetic algorithm, data structures, application, Huffman coding.