Increasing complexity in wireless communication systems—fueled by 5G, 6G, IoT, and autonomous technology requirements—puts increased expectations on the performance, compactness, and agility of antenna systems. Traditional design approaches, i.e., finite element methods (FEM) and full-wave electromagnetic (EM) simulations, while precise, are computational and labor-intensive, particularly with high-dimensional and nonlinear design spaces. These constraints represent significant obstacles to fast prototyping, optimization, and roll-out of novel antenna geometries to accommodate changing demands. To address such demand, machine learning (ML) came forth as an effective enabler of antenna design optimization. Through data-driven methodologies, ML models can learn subtle correlations between antenna geometries and corresponding performance metrics like return loss, gain, bandwidth, radiation efficiency, and directivity. The present work offers a detailed investigation of different ML frameworks, i.e., supervised learning, surrogate modeling, reinforcement learning, and generative design, in the antenna design process. Supervised machine learning algorithms, including deep neural networks (DNNs) and support vector regressors (SVRs), are employed to predict the performance of antennas from geometric characteristics, allowing for fast evaluation without the need for iterative simulations. Surrogate models, generated based on a small number of high-fidelity simulations, enhance design optimization by acting as efficient approximations of computational solvers. Inverse design platforms enable engineers to define some electromagnetic properties and automatically create feasible antenna geometries through machine learning algorithms. Generative models such as Generative Adversarial Networks (GANs) and Variational Autoencoders (VAEs) also possess the ability to find completely new antenna geometries that may elude traditional methods. Reinforcement learning techniques have also been investigated to autonomously control iterative optimizations.