Confused b/w Precision and recall

Course Q&A Machine Learning Specialization Advanced Learning Algorithms
1

I’m getting confused b/w precission and recall
I know the formulas of both but i want any other real life example to understand it in better way.
Could you clarrify ?
thank you!

2

Hi @Raghav_Mahajan Good question. Please check this answer and let me know if this clarifies your question.

1 Like
3

Hi @Raghav_Mahajan,

these metrics are important to understand:

  • false positives (imagine an anomaly detection is sending an alarm even though everything is normal)
  • false negatives (imagine an anomaly detection is sending no alarm even though there is a big anomaly)

Note that you can improve false positives by worsening false negatives and vice versa. So the overall goal is to keep track of both so that they satisfy the business problem in a reasonable accuracy.

Basically: Precision and recall are representing these cases implicitly also have a similar trade off, as shown in this visualization:

878×320 53.5 KB

  • precision is about the relevance of the results provided by the model
  • recall is about the completeness of the results.

They determine the performance of a classification model!

Feel free to take a look at this thread: Anomaly Detection: How to improve? - #2 by Christian_Simonis

Best regards
Christian

1 Like
4

Hello @Raghav_Mahajan,

Besides the great explanations and examples by @pastorsoto and @Christian_Simonis, I would just like to add two more real life examples:

A high precision process is a rather conservative one. We want to apply the strictest standard to make sure the outcome of the process to be always true. For example, a conviction has to be proved beyond reasonable doubt, but because of this high standard, some true cases might get acquitted.

A high recall process is the opposite. For example, when we come across a completely new machine learning topic, we want to search for some articles to inspire us. Since we actually don’t know what’s going to inspire us, we are willing to read perhaps the first 100 searched pages, instead of carefully picking 2 to 3 of them that we know for sure are going to help. However, out of those 100 pages, we might end up spending 2hrs reading 90 of them which turned out to be not helpful, but we are not going to miss the 10 pages that indeed inspired us in 10 different angles.

It is therefore a trade-off process, the more precise result you want, the more true cases you will miss out. The more you want to recall, the more not helpful information you would go through, but the less useful ones you will miss out.

Cheers,
Raymond

1 Like