Thoughts on:

      "Efficient Similarity Search In Sequence Databases",
      Rakesh Agrawal, 
      Christos Faloutsos,
      Arun Swami,
      Foundations of Data Organization And Algorithms (FODO) Conference, 1993

      
This paper presents a method that makes use of Discrete Fourier Transform
to index time sequences for similarity searches. It relies on
the observation that the first few frequencies describe the important
features of the time series. It also relies on Parseval's theorem that
states the Fourier Transform maintains the Euclidean distance in the time
or frequency domain. This technique is applied to sequences of the
same length.
      
This combined with the time warping from Berndt and Clifford might
prove useful as a way of compressing our time series attributes. It 
could be used as a pre-processing step. There still remains the problem
of mining association rules from this type of data set. If performance
problems arise this method of reducing the data might prove very useful
in finding a practical approach.