Best Frag Or K/d

13468912

Comments

  • SacchoSaccho Posts: 1,577Player
    I'd still say the right way to do this is with comparison to expected values.

    Here are some of my stats from one analysis site for another game. For each weapon and each property, a comparison is made against how other players do.
    JLwCwVl.png

    Or consider another similar analytics site, now looking at a 'dirty' player. This is a case where a high headshot rate really jumps out as a glaring inconsistency.
    v6YEXWT.png

    But things like that aren't necessarily going to get you there. This player had fairly normal stats across the board. Low headshot rate, normal frag rate... just a high KDR stood out as odd. But 22 kills over 23 non-headshot hits? That doesn't add up. It's a class of cheese eating designed to get around the kinds of metric-based detections discussed in this topic by bumping up how much damage each bullet does when it lands.
    ndkTD3Z.png



    One thing that hasn't been discussed with your player score system is where the threshold is. It's easy to say "Yeah, a score of 97 is bad". But what about a score of 10? 8? 6.5? Why does your system decide Ali is clean beyond your say-so? I may come off a bit snarky in saying that, but it's a real problem. Some of the YouTube/BF3 stars had to get special exemptions from the stats-based detection systems because they kept coming up as false positives for their high headshot rates.

    Your system is also flawed in that a low accuracy is VERY easy to manipulate. As that term drops just by spamming pointlessly, the player KDR becomes irrelevant. That whole term can go to something like 100 just due to a headshot rate.

    Colts: A*KDR+H = 13.3*4.45 + 41.7 = 100.9
    Hypothetical player: A*KDR+H = 1*25 + 75 = 100

    Suddenly the system is only really considering that player's kills/hour when deciding if he's any better than Colts even though he has a ridiculously high KDR and headshot rate.


    I totally support the endeavor and think you're doing great. I don't mean this post to tear down anything. I just want any system that gets produced to be resistant to getting screwed that way and to make sure everything gets addressed.

    It'll also be much easier to address once an API's officially available and stats can be considered en masse. **wink wink nudge nudge, Devs :) ** Your databases have had it too easy for too long!
  • -=DA=-Tw!$t3d-=DA=-Tw!$t3d Posts: 65Player
    Saccho wrote: »
    I'd still say the right way to do this is with comparison to expected values.

    Here are some of my stats from one analysis site for another game. For each weapon and each property, a comparison is made against how other players do.
    JLwCwVl.png

    Or consider another similar analytics site, now looking at a 'dirty' player. This is a case where a high headshot rate really jumps out as a glaring inconsistency.
    v6YEXWT.png

    But things like that aren't necessarily going to get you there. This player had fairly normal stats across the board. Low headshot rate, normal frag rate... just a high KDR stood out as odd. But 22 kills over 23 non-headshot hits? That doesn't add up. It's a class of cheese eating designed to get around the kinds of metric-based detections discussed in this topic by bumping up how much damage each bullet does when it lands.
    ndkTD3Z.png



    One thing that hasn't been discussed with your player score system is where the threshold is. It's easy to say "Yeah, a score of 97 is bad". But what about a score of 10? 8? 6.5? Why does your system decide Ali is clean beyond your say-so? I may come off a bit snarky in saying that, but it's a real problem. Some of the YouTube/BF3 stars had to get special exemptions from the stats-based detection systems because they kept coming up as false positives for their high headshot rates.

    Your system is also flawed in that a low accuracy is VERY easy to manipulate. As that term drops just by spamming pointlessly, the player KDR becomes irrelevant. That whole term can go to something like 100 just due to a headshot rate.

    Colts: A*KDR+H = 13.3*4.45 + 41.7 = 100.9
    Hypothetical player: A*KDR+H = 1*25 + 75 = 100

    Suddenly the system is only really considering that player's kills/hour when deciding if he's any better than Colts even though he has a ridiculously high KDR and headshot rate.


    I totally support the endeavor and think you're doing great. I don't mean this post to tear down anything. I just want any system that gets produced to be resistant to getting screwed that way and to make sure everything gets addressed.

    It'll also be much easier to address once an API's officially available and stats can be considered en masse. **wink wink nudge nudge, Devs :) ** Your databases have had it too easy for too long!

    I cant seem to find the "Like" button. Bravo.

    Great explanation to highlight my flaws. and a great example for a solution. Thank you sir, i'll stew on this for a bit and see what i can come up with!
  • .sauce.sauce Posts: 308Player
    Correct me if I'm wrong twisted but the snippet you presented shows heads hots counted. The anomaly lies not in counting headshots but in including grenade kills in the percentage. I just did the math on my profile and my headshots/all kills = the headshots percentage even though some of the kills come from grenades. Again I could be wrong.
    Hello sir, excellent accuracy.

    LETS GO PENS
  • -=DA=-Tw!$t3d-=DA=-Tw!$t3d Posts: 65Player
    Ah crap, your right....well this is embarrassing.
  • -=DA=-Tw!$t3d-=DA=-Tw!$t3d Posts: 65Player
    Fixed.

    Headshot percentage is now caculated by taking the
    (Total Headshots / (EnemyNutralized - Grenade Kills) * 100)

    Good Catch, Good Recheck. Thanks for your contribution.
  • -=DA=-Tw!$t3d-=DA=-Tw!$t3d Posts: 65Player
    @Saccho
    Thanks again for your post, I've read into what you've posted and i feel like this is very promising to determine legitimacy, However its going to take a bit to recreate for AAPG, and even more time to test fully. But I am moving forward with this. I'll keep you posted. Thanks for your contribution sir.
  • SacchoSaccho Posts: 1,577Player
    Hey, my pleasure. This little corner of the world is where my loves for gaming and mathematics intersect so I've got some strong opinions built up :smiley: It's easy to quickly throw out "do more math!" in a complaint. Putting it into an app that actually works is the hard part and I respect that effort.

    If you really want to get into it, more complicated approaches can likely do even better (see, for example, some of my older posts on metric analysis for player skill assessment: http://archive2.forum.americasarmy.com/viewtopic.php?p=154683#p154683 and http://archive2.forum.americasarmy.com/viewtopic.php?p=154995#p154995). But at that point it gets complicated fast and really goes to a whole new level.
  • .sauce.sauce Posts: 308Player
    Glad to help. Super cool of you to do all of this in the first place.
    Hello sir, excellent accuracy.

    LETS GO PENS
  • -=DA=-Tw!$t3d-=DA=-Tw!$t3d Posts: 65Player
    .sauce wrote: »
    Glad to help. Super cool of you to do all of this in the first place.

    Lets Go Pens!! (I live just south of da burgh!) Are you a fellow Yinzer?
  • =IK=Doba==IK=Doba= Posts: 2,789Player
    Just saying ..but that hypothetical player of 1*25+75=100.. is not gonna happen

    25 KD 75% headshot with only 1 on accuracy would never happen.. he would have to spam all his bullets away... he would never have anything left to kill anyone with... I get the example it's just unrealistic .. if a player is dirty his accuracy will probably be much higher than average due to programs helping
    _____________________________
    #Support Comp Mode

    https://www.youtube.com/channel/UCN4YhM6jUB2MxVj8i3b9rhw
  • ={GIS}=Nesquick={GIS}=Nesquick Posts: 19Player
    Fixed.

    Headshot percentage is now caculated by taking the
    (Total Headshots / (EnemyNutralized - Grenade Kills) * 100)

    Good Catch, Good Recheck. Thanks for your contribution.

    Same thing for takedown, i think ;)
  • .sauce.sauce Posts: 308Player
    .sauce wrote: »
    Glad to help. Super cool of you to do all of this in the first place.

    Lets Go Pens!! (I live just south of da burgh!) Are you a fellow Yinzer?

    Grew up in FL but I have family in Greensburg, Irwin and other areas. Hope to move close in the next 5 years.
    Hello sir, excellent accuracy.

    LETS GO PENS
  • K!Dz.applePIEK!Dz.applePIE Posts: 1,050Player
    I also agree on accuracy not really being a good constant on players aim.

    when I play inner I always shoot glass at the beginning of rounds.. 1 mag gone for no kill every round..

    I have also the bad habit of spamming upper windows in def as well..

    That being said, I think we really need a formula to flag some fishy players out there, to at least put them on a watch list.
  • =IK=Doba==IK=Doba= Posts: 2,789Player
    If you take all fishy players I guarantee you they'll have an above average accuracy...even with spam..

    You spam and miss shots
    They spam and don't miss shots

    Still results in much higher than average accuracy
    _____________________________
    #Support Comp Mode

    https://www.youtube.com/channel/UCN4YhM6jUB2MxVj8i3b9rhw
  • SacchoSaccho Posts: 1,577Player
    edited October 2015
    I chose examples based on things I've seen real players do to dodge detection. They're not hypotheticals. I've directly observed players in other titles spamming needlessly to game their stats. Hearing a player spamming his LMG nonstop all game long was often a good sign that he needed to be looked into. Just as PunkBuster and other a/c solutions play a cat-and-mouse game, the same happens with statistics-based methods.

    Look at the battle report image I posted. You think it's a coincidence the guy shot exactly 3 times for each 1 hit and that each single hit, despite coming from a low-powered pistol, was a kill? You think he wasn't trying to fake a more believable accuracy number? It's naive to make assumptions about the stats of dirty players and how any particular metric would shake out.

    The specific example wasn't just "1% accuracy would break this", it was "a simple equation is easy to game/beat".

    It goes deeper too.
    Colts: A*KDR+H = 13.3*4.45 + 41.7 = 100.9
    Me, if I were a shotgun-only player: 139*1.9 + 31 = 295
    Lower KDR, lower headshot rate... far higher rating.

    The equation could end up creating false positives for anybody playing only with the shotgun if it didn't account for how that weapon works. There's no way I would ever be rated 3x higher than Colts.

    That's why per-value statistics are the way stats-based a/c systems have gone. It's not just the independent websites, either. Professional services like GameBlocks' FairFight product (notably picked up by EA/Dice for Battlefield 4) are using statistical methods to identify dirty players in the manner I recommend:
    https://gameblocks.com/faqs
    Algorithmic Analysis of Player Statistics (AAPS) compares each player's game play statistics across an array of performance measures and compares them against averages derived from all of the gamers playing your game. FairFight uses the results of these comparisons to find statistical anomalies that correlate strongly to the use of cheats.
  • flubbyflubby Posts: 50Player
    Would be fun to come up with other stats like these:

    "Pink Panty Score": High K/D, low secure & revive ratios, high choked/choke ratio
    "Horny [TOS Violation] Grabber Score": Low K/D, high choke/choked ratio

    :)

    ss_flubby_4.png

  • -=DA=-Tw!$t3d-=DA=-Tw!$t3d Posts: 65Player
    Fixed.

    Headshot percentage is now caculated by taking the
    (Total Headshots / (EnemyNutralized - Grenade Kills) * 100)

    Good Catch, Good Recheck. Thanks for your contribution.

    Same thing for takedown, i think ;)

    Take-downs aren't calculated. Its a stat recorded by the game. Im sure its correct.
    I also agree on accuracy not really being a good constant on players aim.

    when I play inner I always shoot glass at the beginning of rounds.. 1 mag gone for no kill every round..

    I have also the bad habit of spamming upper windows in def as well..

    That being said, I think we really need a formula to flag some fishy players out there, to at least put them on a watch list.

    Welcome to the discussion Kidz. Please read the previous 5 pages, Pretty much every topic of your post has been covered.
    flubby wrote: »
    Would be fun to come up with other stats like these:

    "Pink Panty Score": High K/D, low secure & revive ratios, high choked/choke ratio
    "Horny [TOS Violation] Grabber Score": Low K/D, high choke/choked ratio

    :)

    ss_flubby_4.png

    While i can appreciate the hilarity of what your hinting at here, Right now the focus is on trying to keep players honest.


    Lastly, @Saccho

    You sir, have a great mind, i'd like to discuss this "math" you speak of with you further. If you would please add me on steam i'd really appreciate it. "datwisted718"
  • K!Dz.applePIEK!Dz.applePIE Posts: 1,050Player
    edited October 2015
    Thank you DA I actually read all of the pages (Not to mention I was posting on this thread already before you started) and my post last post was about agreeing with part of your formula being inaccurate by giving another example of spam firing not being part of shooting enemy. But we also established that.. But good luck finding a formula..
  • SSKwaNtedSSKwaNted Posts: 266Player
    edited October 2015
    I think the formula might work just to flag players and then with a check on their full stats you can easily get who's legit and who's not, also the kill x hour might help, but with the database problem, some1 will get impossible stats even if they are legit, but it should be easy to spot those.

    http://www.americasarmy.com/soldier/sskbart

    8 hours 12k kills
    AAPGOD/ Ex R6 Pro
    Achievements
    Youtube
    Steam
  • =IK=SgtBadazz=IK=SgtBadazz Posts: 160Player
    What's the point if everyone's stats aren't correct?
This discussion has been closed.