Active Learning Based on Geographical Orientation for Automatic Transportation Mode Estimation

概要

論文の詳細を見る
We focus on the automatic transportation estimation task, which automatically estimates transportation modes given GPS trajectories of a user. Previous works have used supervised learning frameworks to estimate transportation modes and have reported that it achieves certain performances. However, the main drawback of supervised learning is the requirement of a significant amount of labeled data. Active learning is an effective solution to this problem. Although many studies have developed a wide variety of active learning algorithms, it has previously been unclear as to whether active learning works well for the automatic transportation mode estimation task. In addition, no previous work reveals which aspects are useful for selecting better instances in terms of model improvement for this task. We propose a novel aspect, geographical orientation, to develop a semi-stream-based active learning method. Our method takes into account geographical distance and density separately from the information based solely on feature space.
2014-07-25