Overfitting is arguably the biggest problem in machine learning and data science, and being able to detect it will make you a much better data scientist. While reaching a high (or even perfect) accuracy on training sets is quite easy when you use neural networks, reaching a high accuracy on validation and testing sets is a very different thing.
Let's see if you can now detect overfitting. Amongst the accuracy scores below, which network presents the biggest overfitting problem. ?