Crowd development of a scoring system for the IEM system
Hi, Habr. We have extensive experience in developing systems for enterprise automation, but unfortunately, I almost do not remember anything from mathematical statistics. In general, I had a department of discrete mathematics. But there is an idea to make a project that on the sidelines is called Iron Felix.
It is known that in any company employees are trying to get more benefits than, let's say, it should be. And we would like to realize the search for such non-standard activity. The task is complicated by the fact that we do not know which activity is harmful. But we know that for most users it is normal.
And now we want to find the appropriate specialist for an open project to solve this problem. We want to publish all stages and results here in Habré. Participation in the project is encouraged financially. Actually, the first stage is to formulate the problem in a somewhat correct and structured way. If you feel the strength and interest in such activities, then write in kamentah or in the feedback form on the site, I will contact directly.
UPD.
I will try to clarify the task a little bit. And then there is a feeling that he did not convey the essence in sufficient detail.
')
So, there is an IEM system. This is the next generation of enterprise automation systems development. And for the current task, we can assume that this is just a fresh ERP system, in which all data about the company are consistent, reliable, lie in a single database in a structured form, etc. etc.
Accordingly, there is a lot of information about the actions of the users of the system, the counterparties of their parameters (such as turnover, money flow, profitability, up to a specific SCU unit). There are hundreds and thousands of such parameters.
It is necessary to solve the problem of identifying meaningful parameters, building hypotheses, calculating and searching for appropriate models. As an example, you can bring sales managers. There are 1,000 of them, and they all do roughly the same operations. But someone (maybe) is doing something wrong.
You can, of course, try to look for example those who have more returns to the guarantee, or a small profitability. You can go down to the level of individual operations - sales, purchase of movements and consider the parameters of these individual operations.
But I want to try to make a scheme in which an algorithm will be formulated that will allow to identify significant parameters (or determine that there are none), and then find a method for finding the "suspects" that are somehow different.
What methods to use here is a question for specialists in ML, mathematical statistics and other neural networks.
And actually we are ready to pay for it.
It is known that in any company employees are trying to get more benefits than, let's say, it should be. And we would like to realize the search for such non-standard activity. The task is complicated by the fact that we do not know which activity is harmful. But we know that for most users it is normal.
And now we want to find the appropriate specialist for an open project to solve this problem. We want to publish all stages and results here in Habré. Participation in the project is encouraged financially. Actually, the first stage is to formulate the problem in a somewhat correct and structured way. If you feel the strength and interest in such activities, then write in kamentah or in the feedback form on the site, I will contact directly.
UPD.
I will try to clarify the task a little bit. And then there is a feeling that he did not convey the essence in sufficient detail.
')
So, there is an IEM system. This is the next generation of enterprise automation systems development. And for the current task, we can assume that this is just a fresh ERP system, in which all data about the company are consistent, reliable, lie in a single database in a structured form, etc. etc.
Accordingly, there is a lot of information about the actions of the users of the system, the counterparties of their parameters (such as turnover, money flow, profitability, up to a specific SCU unit). There are hundreds and thousands of such parameters.
It is necessary to solve the problem of identifying meaningful parameters, building hypotheses, calculating and searching for appropriate models. As an example, you can bring sales managers. There are 1,000 of them, and they all do roughly the same operations. But someone (maybe) is doing something wrong.
You can, of course, try to look for example those who have more returns to the guarantee, or a small profitability. You can go down to the level of individual operations - sales, purchase of movements and consider the parameters of these individual operations.
But I want to try to make a scheme in which an algorithm will be formulated that will allow to identify significant parameters (or determine that there are none), and then find a method for finding the "suspects" that are somehow different.
What methods to use here is a question for specialists in ML, mathematical statistics and other neural networks.
And actually we are ready to pay for it.
Source: https://habr.com/ru/post/320598/
All Articles