Mining non-redundant distinguishing subsequence for trip destination forecasting

Trip destination forecasting has received a great attention recently when the idea of intelligent transportation is discussed ubiquitously from related business to government. For instance, we gain convenience such as easily finding a ride at the right location for the right time in a car dispatch system given the sharing economy setting. In this work, we aim to propose a method for trip destination forecasting given its first partial trip trajectory as the input. In a nutshell, we formulate the problem as a multi-class prediction problem and mine the distinguishing pattern that we can see on one class but not on other classes. Moreover, we attempt to find non-redundant rules to separate the interested class from other classes by an efficient algorithm called Non-redundant Contrast Sequence Miner given multiple answers (destination) to choose from. This study tested the proposed method on public trip destination prediction dataset. The results show that the proposed method outperforms other mining techniques on the task of trip destination forecasting in terms of accuracy and resource allocation both time and memory usage efficiency and accuracy.