Harmony Search algorithm for optimal word size in symbolic time series representation

Fast and high quality time series representation is a crucial task in data mining pre-pre-processing. Recent studies have shown that most representation methods based on improving classification accuracy and compress data sets rather than maximize data information. We attempt to improve the number of SAX (time series representation method) word size and alphabet size by searching for the optimal word size. In this paper we propose a new representation algorithm (HSAX) that deals with Harmony Search algorithm (HS) to explore optimal word size (Ws) and alphabet size (a) for SAX time series. Harmony search algorithm is an optimization algorithm that generates randomly solutions (Ws, a) and select two best solutions. H SAX algorithm is developed to maximize information, rather than improve classification accuracy. We have applied HSAX algorithm on some standard time series data sets. We also compare the HSAX with other meta-heuristic GENEBLA and original SAX algorithms The experimental results showed that the HSAX Algorithm compare to SAX manage to generate more word size and achieve less error rates, whereas HSAX compared to GENEBLA the quality of error rate is comparable with the advantage that HSAX generated high number of word and alphabet size.