Forecasting modeling with kernel function integration in gaussian processes

This research proposes to design a kernel within a Gaussian process for finding and learning patterns from data attributes that fit the structure of time series data. No external variables needed. In Gaussian processes, researchers do not need to modify the algorithm's layout at all when the function of the problem converts. So what to do just to modify the function or kernel function to suit the problem. The kernel functions in each of the Gaussian process types affect the different models of time-varying functions. The accuracy of the Gaussian process algorithm depends on the choice of function. Choosing a function of the quadratic function, we will select the corresponding function of the function. Which depends on the pattern of the problem. Selecting and using a single type of Kernel function might not cover and suitable for the problem of respective forecast. Therefore, the different types of Kernel function are combined and new type of Kernel function is generated, thereby, this provides Superposition properties of the Kernel function. This property enables to separately control each type of characteristics of function. This new Kernel function can be used to different problems and patterns under Gaussian process. The Gaussian process doesn’t need to base on the selection of Kernel function and it results in the forecast is more preciously and higher effectively.