Overfitting in ML
From the previous chapters, we now know what overfitting is and its adverse effect when used on unseen data. Let's take a step further by digging into what the root causes of overfitting are, how we can spot overfitting when we build our models, and some important strategies we can apply to curb overfitting. When we gain this understanding, we can go on to build effective and robust ML models.
What triggers overfitting
In Chapter 6, Improving the Model, we saw that by adding more neurons to our hidden layer, our model became too complex. This made our model not only capture the patterns in our data but also the noise in it, leading to overfitting. Another root cause of overfitting is working with insufficient data volume. If our data does not truly capture the full spectrum of variations our model will be faced with upon deployment, when we train our model on such a dataset, it becomes too specialized and fails to generalize when used in the real world...
