Abstract: Online efficient multiple-outputRegression is an important machine learning technique for modeling, predicting,and compressing multi-dimensional correlated data streams. This proposed systemintroduced, a novel online efficient multiple-output Regression method, calledE-MORES, for streaming data. E-MORES can dynamically learn the structure of theRegression coefficients to facilitate the models continuous refinement.Considering that limited expressive ability of Regression models often leadingto residual errors being dependent, E-MORES intends to dynamically learn andleverage the structure of the residual errors to improve the predictionaccuracy. This system also introduce RandomForest and DecisionTree to predict(classify) the next event type that will occur during the transition time, thatis growing, continuing, shrinking, dissolving, merging or splitting. Theconducted experiments suggest that the RandomForest and DecisionTree classifiersusually provide more accurate results. Moreover, system introduce threemodified covariance matrices to extract necessary information from all the seendata for training, and set different weights on samples so as to track the datastreams’ evolving characteristics. Furthermore, an efficient algorithm isdesigned to optimize the proposed objective function, and an efficient onlineeigen value decomposition algorithm is developed for the modified covariancematrix. Finally, this system analyze the convergence of EMORES in certain idealcondition. Experiments carried out on two synthetic datasets and threereal-world datasets validate the effectiveness and efficiency of E-MORES.
Keywords:Decision Tree,Dynamic Relationship Learning, For-getting Factor, Lossless Compression, OnlineEfficient Multiple –Output regression method.