Enhanced minimum description length preprocessing of time series trajectories

Abstract

Mining of moving objects trajectories is becoming more and more popular due to its wide range of real time applications. Discovering knowledge from the time series trajectories is challenging and there is needs to improve the quality of the output generated. There are many effort being made to improve the quality of mining process and also to improve the accuracy of the techniques. Preprocessing of the input raw trajectories play an important role in the mining of trajectories. The objective of preprocessing of input trajectories is to reduce the number of segments of the original trajectories without losing major data from original trajectories. The Minimum Description Length (MDL) principle is used to preprocess the input trajectories and reduce the length of trajectories. There are two major issue with the MDL principle. The first issue faced by MDL, the points which are very close to each other are remained unprocessed. Second issue is, the loss of information using MDL principle is considerably more. There is need to tackle both the problem, so that performance of the preprocessing is improved. In this paper, we have proposed RegressionLine technique to process points of trajectories which MDL principle cannot process. We have proposed RelativeLoss algorithm to preprocess the input trajectories by reducing the loss of data. Proposed techniques were tested on real time and synthetic datasets and showing improvement over existing MDL technique.