Gaussian Process Regression with Average Hyperparameter
Sri Winarni, Sapto Wahyu Indratno · Journal of Physics Conference Series · 2020
Abstract Gaussian regression process (GPR) is a non-parametric analysis approach that is more flexible than simple linear regression analysis. Linear and non linear patterned data can be modeled by GPR. In big data the use of GPR is constrained by the increasing size of the data. Data partitioning approach can be done as an alternative to facilitate data processing. The partitioned data then shifts one data, and so on until the last training data. In each data partition an optimum hyperparameter will be obtained. The average hyperparameter is used in the prediction process. The results obtained are that by partitioning the data and then using the average hyperparameter, the predicted results obtained are better than doing with the overall training data. In the data partition approach, small partitions provide better predictive results than large partitions. The resulting RMSE value is also getting smaller. Thus the average hyperparameter approach with small data partitions can be used as an alternative in making predictions on GPR.