Combining Micro and Macro Features for Online Pattern-of-Life Movement Prediction

We investigate the problem of predicting the future location of mobile objects, such as vehicle, ships, or people, in real-time, with a high degree of accuracy. Our premise is that an effective combination of recent and long-term historical data can significantly improve prediction performance by enabling a representation of the patterns-of-life of the objects. However, finding a feature representation that captures the long-term observed history of the objects in a compact yet informative manner is a key challenge in data mining and machine learning. To this end, we propose to combine "micro" features, which capture recent fine-grained trends, in conjunction with "macro" features, designed to summarize the long-term history in a compact yet meaningful manner. Through extensive empirical studies on marine vessels and people movement data, we evaluate the impact of using both micro and macro features on the learned model and demonstrate that the proposed approach significantly enhances predictive accuracy.