Browse
Category
DiscussionsGeneral

Linear Regression Model for Winning Probability

Closed
kchild90

I've built a linear regression model which predicts the number of entries a giveaway will receive based on:

  • Creation time
  • End time
  • Level
  • Game
    and then using the number of copies can predict the probability of winning a giveaway.

This gives a maybe better / maybe worse idea of the probability of winning a giveaway. For example, this giveaway which has just appeared:
https://www.steamgifts.com/giveaway/Nojef/energia
the steamgifts plugins will all say has a 10% chance of winning, but the model predicts by the time it ends, it will give a 0.19% chance of winning.
Edit: I'm aware it's slightly embarrassing if that is miles off now ;)
2nd Edit: final, actual chance of winning was 0.26%

This is better because it's more realistic, but worse because it could well be quite a way off. I've not tested the model at all extensively yet.

Thoughts on this? Is there any interest in it? I have wondered about an implementation which allows you to cut out giveaways with a predicted chance of winning less than a certain amount, as you can currently with current chance of winning.

7 years ago*

Comment has been collapsed.

7 Comments
Oppenh4imer

Could you publish your model, so we can take a look.

7 years ago
Permalink

Comment has been collapsed.
kchild90

Potentially. It would just be the coefficients from r. The variable "game" has over 1,000 different levels though.

7 years ago
Permalink

Comment has been collapsed.
Oppenh4imer

Do you take timezones into account for times?

Is there a difference between public and private GAs?

7 years ago
Permalink

Comment has been collapsed.
kchild90

Yeah. One of the things I was most interested in was whether you're better off entering giveaways that end at a certain time. Similar to how on ebay if you enter a local auction which ends at 3AM local time there will be less interest in the item. Still, gonna delete this now - see Lilith's post below for why :)

7 years ago
Permalink

Comment has been collapsed.
Lilith812

Well, I don't know who would need this: people who choose carefully giveaways to enter would anyway enter for games they would play, even if chance is small, and they have always a lot of points (300 almost all the time).
Is it for bots to skip GAs with very low chance to win? Make botting even more profitable..

By the way, what dataset do you use? Is there available dataset in open access? Or you picked manually some data? How many dataitems do you have?

7 years ago
Permalink

Comment has been collapsed.
kchild90

Fair enough. My thought was I often have quite low points and end up entering whatever is finishing soon and still has a high chance, on the grounds that the chance can't reduce too much in the final hour.

Don't want to assist with botting though so will not take this further. Partly just did it as a project of personal interest.

7 years ago
Permalink

Comment has been collapsed.
AllTracTurbo

Good luck predicting the number of entries on a group giveaway with a large group compared to a level 0 public giveaway :)

https://www.steamgifts.com/discussion/2zLgH/nothing-fishy-going-on-here

7 years ago
Permalink

Comment has been collapsed.

Closed 7 years ago by kchild90.