Reaction kinetics of chemical ionization mass spectrometry (CI-MS) based ion-molecule reactions is an important component in the quantification of trace-level volatile organic compounds (VOCs). The rate coefficients of such CI-MS reactions are predicted using the Gaussian process regression (GPR) machine learning method from the dipole moment, polarizability, and molecular weight of the molecules, mitigating experimental complexity in CI-MS rate coefficient estimation. GPR can make predictions combining prior knowledge (kernel function) which is considered the heart of the GPR model and provide uncertainty measures over predictions. A suitable kernel combination with proper tuning of parameters can make the Gaussian process more robust and powerful. Various kernel combinations, such as kernel addition and multiplication, are tested in the GPR prediction of rates. A blend of radial basis function (RBF), white noise, and squared exponential kernel performs better, and the predicted rates are in close agreement with the experimental rates. GPR provides an alternative to the capture collision rates and can be useful when there are no experimental data available and/or the available data contain large uncertainty in the rate coefficients.