This chapter explores the evolution of freeway traffic modeling, highlighting the shift from traditional physics-based approaches to modern data-driven methods. Historically, traffic flow models were grounded in fundamental principles of physics, such as the conservation of vehicles, the dynamics of their interactions with other vehicles and the road infrastructure, and the vehicle motion equations. These models have been proven to be quite useful in predicting traffic conditions in freeway networks but sometimes struggle to capture the complexities and variability inherent in real-world traffic systems. With the advent of big data and machine learning techniques, a new paradigm has emerged, i.e., purely data-based models or physics-informed machine learning models, the latter incorporating physical laws and domain-specific knowledge directly into the learning process. In this chapter we will present a literature review on these different traffic modeling approaches, and we will introduce an innovative physics-based machine learning modeling framework, based on a physics-regularized Gaussian process model and a physics-informed long short-term memory network. This framework has been tested on real traffic data referred to the I-8 Kumeyaay Highway in California and has been compared with purely physics-based and purely data-driven approaches, as discussed at the end of this chapter.