PMEM: Predicting multiple time series using ensemble model
Vishwanath R Hulipalled, Lalit Mohan Patnaik, K. C. Srikantaiah, K R Venugopal · 2016
Forecasting Multiple Time Series (MTS) consists of multiple time series with no relation between them and independent of each other. Predicting each time series independently may lead to increase in time and cost. In this paper, we formalize the problem of predicting the multiple time series together over a MTS database. The proposed framework addresses the following issues. First, it build the initial ensemble model for each time series by using a novel Ensemble approach thereby effectively reduce the data storage and time complexity, secondly, by using single ensemble engine for MTS we perform the three major task, i.e., the task of prediction, building new model, ensemble update and lastly predicting the samples by pattern sequence matching using well known Sliding window method. The computational cost for PMEM is O(C * N * (K+ nmatches)) where, nmatches is the Average number of patterns matched and the storage cost is O (N * (\Di\ + C)). The experimental results show the effectiveness of our approach in predicting the multiple time series data compared with the predicting the single time series data using existing methods.