Register Now

Login

Lost Password

Lost your password? Please enter your email address. You will receive a link and will create a new password via email.

Login

Register Now

It will take less than 1 minute to register for lifetime. Bonus Tip - We don't send OTP to your email id Make Sure to use your own email id for free books and giveaways

Confusion Matrix in Data Science, meaning and example

What is Confusion Matrix?
Confusion Matrix is a performance measuring technique for ML Classification model.

Why do we need Confusion Matrix? Is measuring accuracy not enough?
Confusion Matrix suggests the actual accuracy of your model. For example. Suppose I want to classify if a person is suffering from a very rare disease(1/100000). Even If i build a very bad model and label everyone as ‘suffering’ from the disease, then also the accuracy of the model will be somewhere around 99%. But that model is of no use because it is unable to solve the problem of classification.
Here comes the confusion matrix which is a 2×2 matrix of predicted and actual values

Here the columns denote the Predicted values and rows denoted the Actual values.

Take example of a fire alarm
True Positive – Prediction is true i.e. there is a fire in the building and there is actually fire in the building. That’s fine
False Negative – There is ‘actual’ fire in the building but your model suggests that the alarm is ringing in vain. This is catastrophic, the same is with the disease example, i.e. the person is infected but the model is unable to identify. It’s type II error
False Positive – The building is not on fire but the model suggests that it is on fire. This is still acceptable, matlab kaam chal jaeyga. Example. The person is not infected by the virus but your model suggests that it is, you will go for a few check-ups and will confirm that you are safe :). This is Type I error
True Negative – No fire, no alarm – All chill

Accuracy = (TP+TN)/(TP+TN+FP+TN)
Precision = TP/(TP+FP) i.e. the accuracy of the positive prediction
Recall Sensitivity = TP/(TP+FN) i.e. Coverage of actual positive results
Specificity = TN/(TN+FP) i.e Coverage of actual negative results

When is precision more important than recall?
Suppose there is a Zombie apocalypse, in that case you want to put as many normal person as possible. But even a single infected person is dangerous, so you look for high precision i.e. less False positive cases

What do they ask in top Data Science Interviews: 5 Complete Data Science Real Interviews Q and A
What do they ask in Top Data Science Interview Part 2: Amazon, Accenture, Sapient, Deloitte, and BookMyShow

Keep Learning 🙂

The Data Monk

About TheDataMonkNewbie

I am the Co-Founder of The Data Monk. I have a total of 4+ years of analytics experience with 3+ years at Mu Sigma and 1 year at OYO. I am an active trader and a logically sarcastic idiot :)

Follow Me

Comment ( 1 )

  1. Great content! Super high-quality! Keep it up! 🙂

Leave a reply