When we perform supervised machine learning we often try different algorithms and produce multiple data models. At that point we can either choose the best one or combine several into an ensemble model. In order to do either of these effectively we need to evaluate the models: this we can do in a variety of ways. One of the best tools we have is the Receiver Operating Characteristics (ROC) curve. However, like all tools, ROC curves can be used or abused. This talk will introduce them, highlight their strengths and also their weaknesses so that you can use them effectively.
You can see Mark’s slides via the link below: