Confusion Matrix — Explained via Rain & an Umbrella
Ron’s job is to keep people informed about inclement weather, so that they can plan their days better. Everyday, Ron wakes up, observes the surroundings, and predicts if it will rain
Before people leave for work, they turn on the radio to listen to Ron’s prediction. They carry an umbrella based on what he predicts
Ron meets his nemesis
One day, Ron was challenged by a guy named Brick
Brick said : “Ron, I challenge you to a duel, whoever is better at predicting rain will win”
Ron accepted, obviously
Taking stock of reality
In real world, there are only 2 outcomes possible (think Classification problems)
- It rained
- It didn’t rain
Let’s visualise this as a 2x2 quadrant, where we map both what happens in reality and what Ron predicts
There are a total of 4 scenarios
- Ron said “it will rain”, and it rained — HIT (True Positive)
- Ron said “it will not rain”, and it rained — MISS (False Negative)
- Ron said “it will rain”, and it didn’t rain — FALSE ALARM (False Positive)
- Ron said “it will not rain”, and it didn’t rain — CORRECT REJECTION (True Negative)
Duel time
To find the winner, people started asking the question :
On all the days that it rained, how many days did Ron get right
Hit Rate = ( No of times Ron said it will rain / No of times it really rained )
Based on the observation below, Ron’s Hit Rate = 2/5 = 0.4, which means on all the 5 days that it rained, Ron could only get 2 days right
Brick’s Hit Rate = 2/5 = 0.4
Hit rate is also known as Recall OR Sensitivity OR Statistical Power
Both Ron and Brick ended up with the same Hit Rate. Interesting!
Tie breaker
Ron & Brick’s equal Hit Rate becomes talk of the town. Then a good folk had a smart moment.
Said she : “Yesterday, Ron had predicted that it would rain, but it didn’t! I had to carry an umbrella unnecessarily. Shouldn’t we also check how many times did they raise a false alarm and made us carry umbrellas for no reason?”
It made sense, because no one likes carrying an umbrella unnecessarily
So now, they all started asking one more question :
On all the days that Ron said it will rain, did it actually rain
Precision = ( No of times it rained/No of times Ron predicted it will rain)
Ron’s Precision = 2/5 = 0.4
Brick’s Precision = (2/3) = 0.66
So, who’s better at predicting rain?
It depends…
Imagine the following scenarios
- Ron keeps saying it won’t rain, but it rains. As a result, you get wet more often
- Ron keeps saying it will rain, but it doesn’t rain. As a result, you often end up carrying an umbrella unnecessarily (you won’t get wet though)
Which scenario would you prefer?
Scenario 1:
You are ok with getting wet more often
You are ok with Ron Missing more often
You don’t care so much about Hit Rate
You care about Precision
Scenario 2:
You are ok with frequently carrying an umbrella for no reason
You are ok with Ron raising False Alarms more often
You don’t care so much about Precision
You care about Recall
Of-course the above is an oversimplification. There are instances where you would care both about Precision and Recall. That, is however outside the scope of this article.
Read this excellent article which inspired this verbose post, and thanks for reading! Leaving feedback is strongly encouraged…