Agile Prediction of Ongoing Temporal Sequences Based on Dominative Random Subsequences

This paper identifies a new paradigm of prediction, Agile Prediction of ongoing temporal sequences, which achieves an acceptable accuracy just by the historical subsequences as short as possible and as close to the predicted time point as possible. To address agile prediction, a new concept, Dominative Random Subsequence (DRS for short), is first introduced to capture the local influence and local regularity of the subsequences that are decisive to the future of an ongoing temporal sequence. DRS mining algorithm, MDRS, and its optimal implementation OptMDRS, are also presented. In MDRS and OptMDRS, DRSs are organized as a suffix tree, DRS-Tree, to facilitate the retrieval. Next, this paper proposes an agile prediction algorithm, AgilePredict, to make accurate predictions based the DRS that is closest to the predicted time point. Finally, the results of the extensive experiments conducted on synthetic and real data sets show that our proposed method is feasible and efficient for agile prediction.