Comparison of artificial neural networks and support vector machine methods for computing resonant frequency of rectangular microstrip antenna

In this thesis, by help of High Frequency Structural Simulator(HFFS) software, Rectangular Microstrip Antenna(RMSA) with coaxial fed is designed, and length (L), Width (W), dielectric substrate height (h), and the dielectric constant of substrate (ɛr) as input parameters are used to obtain resonance frequency (fr) and to generate 210 data cluster. With this data cluster, models designed for Artificial Neural Network and Support Vector Machine, which are artificial intelligence methods, are trained and tested, and then by comparing the errors that obtained from these two methods, it is studied on determining the which method yields the best accurate results. In litreature as far as we can examine, in the papers that are use ANN and SVM algorithm for computing resonant frequency of MSA, it is recognized that cross validation approximation providing the reliability of designed model is not used, and therefore using the applying cross validation has been also a target of the of this search. At the first stage, ANN and SVM models are trained with 180 pieces of 210 data generated for RMSA and the remaining 30 pieces are used for testing these trained models, and average percentage error (APE) are calculated. Then, to subject data cluster for cross-validation, 210 data are shifted by 30, and training and testing stages done at the first stage are repeated. This shifting in 210 data is repated by 7 times and for each stage, a new APE value is obtained. Finally, Cross Validation Average Percentage Error (CVAPE) is obtained from all seven APE values. The obtained results for ANN model show that lowest APE (%) and lowest CVAPE (%) are calculated as 0,271 and 0,510 respectively. On the other hand for SVM model lowest APE (%) and lowest CVAPE (%) are calculated as 0,319 and 0,791 respectively. Keywords: High Frequency Structural Simulator, Microstrip Patch Antennas, Artificial Intelligence Methods, Artificial Neural Networks, Support Vector Machines