TABLE OF CONTENTS Indexing of time-series has concentrated great attention in the research community, especially in the last decade, and this can partly be attributed to the explosion of database sizes. Characteristic examples are environmental data collected on a daily basis or satellite image databases of the earth [1]. If we would like to allow the user to explore these vast databases, we should organize the data in such a way, so that one can retrieve accurately and efficiently the data of interest.
More specifically, our objective is the automatic classification of time-series using Nearest Neighbor Classification (NNC). In NNC the time-series query is classified according to the majority of its nearest neighbors. The NNC is conceptually a simple technique, but provides very good results in practical situations. This classification technique is particularly suited for our setting, since we are going to use a non-metric distance function. Furthermore, the technique has good theoretical properties; it has been shown that the one nearest neighbor rule has asymptotic error rate that is at most twice the Bayes error rate [13].
So, the problem we consider is: "Given a database 
of time-series and aquery 
(not already in the database), find the sequence 
that is closest to 
." We need to define the following:
A realistic distance function that will match the user's perception of what is considered similar.
An efficient indexing scheme, which will speed up the user queries.
We will briefly discuss some issues associated with these...
