A Strategy to Improve Accuracy of Multi-dimensional Feature Forecasting in Big Data Stream Computing Environments

High accuracy of multi-dimensional feature forecasting is very important for online scheduling in big data stream computing environments. Currently, most stream computing systems only consider historical features, with future features ignored. In this paper, a strategy to improve accuracy of multi-dimensional feature forecasting for online data stream is proposed. It includes the following contributions. (1) Profiling principles of accurate future feature forecasting objectives from multi-dimensional big data streams. (2) Extracting future features from multi-dimensional historical features of data stream via an improved hybrid IGA-BP (Immune Genetic Algorithm and Back Propagation) algorithm. (3) Evaluating accuracy of future feature forecasting and acceptable latency objectives in big data stream computing environments. Experimental results conclusively demonstrate the efficiency and effectiveness of the proposed strategy.