A hybrid kernel: Fast-Fourier transform kernel model

In reality, most relationships between random variables are nonlinear, and forcing a linear fit can lead to inaccurate predictions and poor model performance. Linear models are unsuitable for most situations due to a lack of flexibility to capture complex patterns and issues with overfitting when data is complex and underfitting when the model is too simple to capture the underlying pattern in the data. Thus, this study compares the performance of the proposed Fast Fourier transform kernel regression with some nonlinear models using simulated and real-life data. The result shows that integrating kernel regression with the Fast Fourier transform proves to be an efficient model based on a minimum root mean square error value of 0.341 and a maximum coefficient of determination (R2 ) value of 0.81 for simulated data, while real-life data had values of 0.007 and 0.94, respectively.