Comparative Regression Analysis for Estimating Resonant Frequency of C-Like Patch Antennas
Umut Özkaya, Enes Yi̇ği̇t, Levent Seyfi, Şaban Öztürk, Dilbag Singh
Abstract
This study provides a comparative analysis of regression techniques to estimate the operating frequency of the C-like microstrip antenna. The performance of well-known regression techniques such as linear regression (LR), regression tree (RT), support vector regression (SVR), Gaussian regression (GR), and artificial neural network (ANN) is tested. For this purpose, 160 C-like microstrip antennas are simulated, of which 145 are used for training of regression techniques and 15 for testing. From the evaluated results, it is found that the pure quadratic Gaussian regression (PQGR) technique has the lowest error rates with 0.0109 mean absolute error (MAE), 0.0087 median error (ME), 0.0002 mean squared error (MSE), 0.0156 root mean squared error (RMSE), and 0.5981 average percentage error (APE). As can be seen in the comparative analysis, the PQGR method outperforms other regression methods on simulation and measurement data. Experimental analysis shows that the resonant frequency of the C-like patch antennas can be calculated very close to measurements.