<![CDATA[The problem of this research stems from the low accuracy of conventional temperature prediction models in representing nonlinear patterns of complex and uncertain climate data. In addition, the utilization of kernel-based machine learning models is still not optimal in improving the accuracy of monthly temperature predictions. The main objective of this study is to analyze, evaluate, and validate the accuracy level of the Bayesian Relevance Vector Machine (BP-RVM) model with a combination of Radial Basis Function and Polynomial kernels in predicting monthly temperatures. This research is an experimental study with a training-testing design. The research subjects are monthly temperature data from Mataram City for the period 2013–2023, with 578 data as test subjects and data from 2023 as testing data. Data collection was carried out through secondary data documentation from NASA, with instruments in the form of temperature datasets processed using MATLAB. Data analysis uses Mean Squared Error (MSE) and Mean Absolute Percentage Error (MAPE). The results show that the BP-RVM model with the RBF-Polynomial kernel, especially with the trainrp algorithm, produces higher prediction accuracy than trainlm. The study's conclusions confirm that the combination of kernels in BP-RVM effectively improves the accuracy of temperature predictions. The implications of this research support the development of more reliable climate prediction models for the agriculture, energy, and tourism sectors.]]>
Copyrights © 2025
