Farkli Örnekleme Yöntemleri̇ni̇n Yapay Si̇ni̇r Ağlarinda Performansinin Karşilaştirilmasi

Abstract

In many modeling problems, the selection of the data to be used is of great importance. In this study, the performances of different sampling methods for artificial neural network modeling were compared on the microstrip patch antenna by using a multilayer perceptron, which is the model that is frequently used in antenna designs. The selected artificial neural network model consists of 5 input and 1 output parameters. The application of Latin Hypercube and Monte Carlo sampling method in the modeling problem for microstrip patch antenna design is investigated. First of all, training and test data sets are provided according to their unique creation method for Latin Hypercube and Monte Carlo examples. In the performance comparison, a total of 12 different networks with 3 algorithms and 4 different architectural structures and 4 different training and test data sets were used. When the results were compared and analyzed with each other, it was seen that the Monte Carlo sampling method gave more successful results in terms of performance in both small and large sample numbers. On the other hand, it was seen that the Latin Hypercube sampling method, on the other hand, increased the number of samples and caused partial improvement. However, it still lags behind the other Monte Carlo sampling method, which has less sample size, in terms of performance. Therefore, it was concluded that Monte Carlo sampling method is more applicable for this problem.