Parameters Selection for SVR Based on the SCEM-UA Algorithm and its Application on Monthly Runoff Prediction

Support Vector Machines (SVMs) have become one of the most popular methods in Machine Learning during the last few years, but its performance mainly depends on the selection of optimal parameters which is very complex. In this study, the SCEM-UA algorithm developed by Vrugt is employed for parameters selection of Support Vector Regression (SVR). The SCEM-UA algorithm, which operates by merging the strengths of the Metropolis algorithm, controlled random search, competitive evolution, and complex shuffling, can avoid the tendency of falling into local minima. The proposed method was tested on a complicated nonlinearly runoff forecasting. The results illustrated that SCEM-UA algorithm can successfully identify the optimal parameters of SVR than grid search method, and can achieve an accurate prediction. Keywords: Support Vector Machines; Optimization; SCEM-UA; Time series; Forecasting

[1]  Vladimir Brusic,et al.  Prediction of MHC class II-binding peptides using an evolutionary algorithm and artificial neural network , 1998, Bioinform..

[2]  D. Rubin,et al.  Inference from Iterative Simulation Using Multiple Sequences , 1992 .

[3]  S. Sorooshian,et al.  A Shuffled Complex Evolution Metropolis algorithm for optimization and uncertainty assessment of hydrologic model parameters , 2002 .

[4]  D. Zaller,et al.  Prediction of peptide affinity to HLA DRB1*0401. , 1994, International archives of allergy and immunology.

[5]  Christian Igel,et al.  Evolutionary tuning of multiple SVM parameters , 2005, ESANN.

[6]  Kalyanmoy Deb,et al.  A fast and elitist multiobjective genetic algorithm: NSGA-II , 2002, IEEE Trans. Evol. Comput..

[7]  H. Kalbacher,et al.  Characterization of peptides bound to extracellular and intracellular HLA-DR1 molecules. , 1993, Human immunology.

[8]  Francis Eng Hock Tay,et al.  Support vector machine with adaptive parameters in financial time series forecasting , 2003, IEEE Trans. Neural Networks.

[9]  Arne Elofsson,et al.  Prediction of MHC class I binding peptides, using SVMHC , 2002, BMC Bioinformatics.

[10]  E. Huarte,et al.  Specific and general HLA-DR binding motifs: comparison of algorithms. , 2000, Human immunology.

[11]  X. C. Guo,et al.  PSO-Based Hyper-Parameters Selection for LS-SVM Classifiers , 2006, ICONIP.

[12]  Miles P. Davenport,et al.  An empirical method for the prediction of T-cell epitopes , 2004, Immunogenetics.

[13]  R. Germain,et al.  Antigen presentation by MHC class II molecules: invariant chain function, protein trafficking, and the molecular basis of diverse determinant capture. , 1997, Human immunology.

[14]  Hans-Georg Rammensee,et al.  MHC ligands and peptide motifs: first listing , 2004, Immunogenetics.

[15]  Z. Nagy,et al.  Precise prediction of major histocompatibility complex class II-peptide interaction based on peptide side chain scanning , 1994, The Journal of experimental medicine.

[16]  Vladimir N. Vapnik,et al.  The Nature of Statistical Learning Theory , 2000, Statistics for Engineering and Information Science.