New Kernel Design Technique was presented in the form of the sum of Linear Kernel, the multiplication of 3 Kernel Functions, including Squared Exponential Kernel, Linear Kernel and Rational Quadratic Kernel, and the multiplication of 2 Kernel Functions, including Periodic Kernel and Linear Kernel, which was used as a component for finding answers in the Gaussian processes. The results showed that the mean absolute percentage error predicted by the New Kernel Function, when the sample size was 180 (8.50E-15), was lower than the Squared Exponential Kernel (SE), Periodic Kernel (PER), Rational Quadratic Kernel (RQ) and Linear Kernel (LIN), which gave the Average Absolute Error of 2.57E-07, 5.56E-02, 2.63E-08 and 4.35E-01, respectively.
SukonthipSuphachan, Poonpong Suksawang, Jatupat Mekparyup (2017); New Kernel Function in Gaussian Processes Model;
Int J Sci Res Publ 7(7) (ISSN: 2250-3153). http://www.ijsrp.org/research-paper-0717.php?rp=P676589