Abstarct: In this thesis, a method for determining the optimal geometrical characteristics of log-periodic dipole antennas using a machine learning algorithm has been presented. Log-periodic dipole antennas typically have multiple geometrical characteristics, and their design using traditional methods is very difficult and costly. The goal of this thesis is to obtain the optimal geometrical characteristics for the log-periodic dipole antenna. For this purpose, the capabilities of neural networks in approximating nonlinear and complex functions have been utilized, and the antenna response characteristics including VSWR, gain, and front-to-back ratio have been estimated as functions of their geometrical characteristics. In this thesis, three parallel and independent neural networks have been used for three different performance characteristics. Moreover, the neural network system has been trained using data from the analysis of two antennas as training data by CST software. The design frequencies of these two antennas are 350 to 1800 MHz and 250 to 720 MHz, respectively. In the optimization process, by defining an appropriate objective function and optimizing it with the Particle Swarm Optimization (PSO) algorithm, the values of the optimal geometrical characteristics of the log-periodic dipole antenna have been determined to achieve the defined objective function. Comparing the results of this method with the corresponding values obtained from characteristic analysis or common optimizations in previous studies indicates the accuracy of the proposed method. This method provides a systematic technique for optimization that is much faster than conventional methods. The reason for this is that in the proposed method, the CST program, which is time-consuming to run, is used only a limited number of times solely to generate training data, and the optimization is performed in the MATLAB software environment with the Particle Swarm Optimization algorithm without the need to run CST.