Simple ranking/rating algorithm

1.1k Views Asked by At

Is there a Simple ranking/rating algorithm that calculates a score between 0 and 1 given a number of alerts along with its priority.

For example: each system can produce "Low", "Medium" and "High" alerts. I have 100 systems to monitor. How can I create a score based on the number of alerts?

Example: System-1 has 10 Medium and 10 High issues. System-2 has 15 High issues. I need System-2 to rank higher.

Why algorithm? - Alert priority is just one parameter. I have 3 other parameters. So I need an algorithm to compute a score based on an alerts parameters.

Please advice...

2

There are 2 best solutions below

0
On

Based on what you told us so far, there are infinitely many scores you can assign.

For example, you could say that a low alert gives $1$ point, a medium alert gives $2$ and a high alert gives $3$.

Then, sum all the points you get from alerts, and divide by $300$ (the maximum possible number of points you can get).


But the $1,2,3$ in my example are completely arbitrary. You could equally have $1,10,100$, but I don't know which one fits best for you...

0
On

Use Logistic Regression with all alert parameters as feature vectors and it will give a score between 0 and 1. Please read about Logistic Regression formulation. If you are new to mathematics, it might be a little difficult but if you give your alert parameters, I shall formulate one and send it to you by email. To read upon logistic regression. https://en.wikipedia.org/wiki/Logistic_regression. The y-label that you pass is something like 1) Action required 2) Action not required and make it as a binary classification. Then all your alert paramaters could result in one of the above mentioned class. The scored probabilities will give you a rank between 0 and 1. Really cool. I am currently doing this great stuff at work. Works very neatly. You could have as much as 100 alert parameters and shall work readily. If you need further help, give me the data for 100 systems and the class they belong from the past historical data that you took action or not. This is just an idea.

Good luck

Satish