Now regularization as it applies to machine learning and data science is the approach for resolving problems with overfitting by discouraging complexity.

And if you're not familiar with what overfitting is, at a high-level overfitting is when you try to pack too much data into your learning model.

So for an example right here

imagine that you had this machine learning algorithm. And in the learning model, you had features such as year, make, mileage, fuel, repairs, and services. And then as time went on you started to realize that the model was not giving the best predictions and what could be the case many times is that you have packed too many data elements and certain classes and certain features are dominating.

So, for example, the year or the fuel feature may be dominating the prediction and that is overfitting. And so what we can do with regularisation is we analyze the different data points and then we remove the elements that we feel are not needed and we only keep the ones that we feel are going to help make the best predictions.

So for this one example we could use regularization to update the training model and only include mileage, repairs, and services. And so we've pruned off those other items from our data set and we've determined using regularization that these are the elements that are going to help give us the most accurate prediction and it's going to help fix our issue with overfitting and it does that by reducing complexity and when we say reducing complexity usually we mean reducing the number of features that go into the training models set.