A model is the reproduction of some aspect from the real world. Good models are parsimonious, so they focus on the parts that are relevant. In machine learning, models also always reflect some aspect from the real world. To be more precise, they usually try to predict or determine a certain target variable by using input variables.

A machine learning model consists of mathematical functions (i.e. a set of rules) that are supposed to reflect the underlying connection between the target variable and the input variable. Machine learning models are developed through so-called model training.

There are different model classes in the field of machine learning that differ in how they represent the real world. Depending on the underlying problem, some model classes are more suitable for certain processes than others. 

Machine learning models are at the core of the vast majority of AI systems today. 

Machine learning models are developed through a process called model training. During model training, we provide a machine learning model with existing data from which the model can identify and show the relevant relationships between independent and output variables.

In our height-weight example, the model was trained using a dataset of a few dozen people. When we want to train a model, we have to decide on a model class. There are different model classes in machine learning; for example, linear models, tree-based models and neural networks.

These model classes differ in how they represent the real world. Therefore, certain classes of models are particularly suited to certain problems. For example, neural networks try to mirror the structure of the human brain to reflect real-world processes and are particularly well suited to computer vision and natural language problems.