Definition Hyperparameter

In the context of modeling, a hyperparameter is basically a variable for which an assumption is taken.

If, for instance, we want to predict the next COVID-19 wave in a certain country during the COVID-19 pandemic, in our model we might wish to assume that a person infected with COVID-19 is contagious for 15 days. This would be the hyperparameter in our model, and ideally we would recalculate our model under different settings of this hyperparameter (what if an infected person is contagious for 5 days, 10, 20, or 25 days?).

In machine learning, hyperparameters are usually set up before the learning process begins (or, in other words, before the model begins to make use of stored observations to improve its predictions). Within the learning process, hyperparameters are usually adjusted to improve the model's performance.