From the course: Mistakes to Avoid in Machine Learning

Unlock the full course today

Join today to access over 22,600 courses taught by industry experts or purchase this course individually.

Evaluating by accuracy alone

Evaluating by accuracy alone

From the course: Mistakes to Avoid in Machine Learning

Start my 1-month free trial

Evaluating by accuracy alone

“

- When it comes to machine learning, we're inclined to think that our prediction should be accurate, above all, but is accuracy enough? In some cases, only evaluating your model by its accuracy can fool us into thinking we have a great model when we actually don't. So in machine learning, accuracy is really just the share of all total predictions that were correct. Now let's consider an example. Now, let's say you are looking to predict the occurrence of bank fraud in your data, you have a set of labeled training data, and in 5% of records, we've identified fraud. This means the remaining observations are not fraudulent. Now, what if I told you I could create a predictive model on this data that was 95% accurate? Sounds pretty good, right? Well, to do this, I could simply always predict no bank fraud, and sure enough, I'd be right 95% of the time. This example of an imbalanced classification problem is a good motivator…

Contents