Research Article

Multivariate and Online Prediction of Closing Price Using Kernel Adaptive Filtering

Table 2

Parameter description for close price using ten different KAF algorithms.

ParameterKAPAKLMSKMCCKNLMSKRLSLKAPALMSNORMAPROB-LMSQKLMS

4.04.04.04.03.05.07.03.0
1.71.11.51.70.091.11.2
1E-41E-20.3
1E-41E-20.4
2
6
mu00.220.2
(P)2020
nu1E-2
500
tcoff0.9

 = kernel width,  = variance of observation noise,  = variance of filter weight diffusion,  = step size,  = regularization parameter,  = Tikhonov regularization, tcoff = learning rate coefficient,  = memory size (terms retained in truncation), mu0 = coherence criterion threshold, P = memory length, and nu = approximate linear dependency (ALD) threshold.