Personal tools
You are here: Home DotA Statistics for the DotA heroes
Document Actions

Statistics for the DotA heroes

by François Rioult last modified 2008-08-23 22:01

Analysis of 181 games 5v5, from DotA 6.44 to 6.50b Also avalaible on


181 5v5 games played on the internet have been analysed, in order to compute statistical indicators about the performances of the DotA's heroes and their equipment. These games have been realized under various conditions:
  • public on the server
  • in a clan, sometimes with public gamers
  • in private,  during games organized by the Dota-League server.
No game has been played in easy mode.


The image of the final board game is transformed in XML data with the help of OCR techniques. These data are stored on a XML database repository eXist and queried with the Xquery language. The frequently chosen items are computed with data mining techniques. The results are render with XSL transformations.


Many average indicators have been computed, with relation to the 181 games of the corpus.
  • the ratio of won / lost games (WinLose)
  • the average of the ratio kill / death (and not the ration average kill / average death) (KillDeath)
  • the number of time the hero kills another hero (kill)
  • the number of time the hero dies (death)
  • le number of creep kills by progression level (creepKill / Lvl)
  • the number of creep denies by progression level (creepDeny / Lvl)

For every hero, a score has been built with the help of the average indicators. It is as higher as a hero kills many other heroes, creeps and dies few. In order to not emphasize the creep farmers, the exact formula is:
score = kill / death * log(creepKill)

The complete table can be viewed at this address:

Click on the colum headers to change the sort order.

CreepKills / Lvl
CreepDeny / Lvl


Even if the indicators are arguable, they are easy to compute and give already interesting results:
  • NerubianWeaver has the best score, because it is a very balanced hero that farms a lot a creeps. It is followed by BloodSeeker and Puck, some awesome killers. The worst scores come to Lanaya and Naix new version, that kill few heroes.
  • The less dying hero are Panda, whose ultimate allows him to go out of every situation, and Warchief, having great strength and stun available.
  • The hero dying at most is without surprise Techies with his kamikaze ability, then Meepo, that has more chance to die than others.
  • The best creep killers are without contest Visage and its revenants, Venomancer and its snakes, NerubianWeaver and its double hit, and Magnataur with its wave and splash.
The frequently chosen items confirm also the recommendation of the experts :
  • Aghanim's Scepter for the heroes whose ultimate is bettered with: Luna, Leshrac, Lich, WitchDoctor, Rhasta, Furion, Ogre, Lion, Zeus
  • ArcaneRing fort the heroes that have to cast a lot of spell to be efficient: EarthShaker, Lina, OmniKnight, Techies, Zeus
  • the teleportation dagger for the heroes that can stun: : Crixalis, Leviathan, Magnataur, MogulKahn, WarChief.
  • almost all hero buy teleportation boots.


This study has of course its limits:
  • only 181 games has been parsed. This is few, but give already interesting results and confirm some well known facts. In average, each heros has been picked 15 times, it is enough to get primary knowledge.
  • These games have been played mostly in dota-league, but there are some public or clan games. The study should differentiate the experimental conditions and lead to different results.
  • The used techniques allows to repeat the entire process with much more results. In fact, the described process is a part of an under development website, where players can upload their games, arrange them in folders (to diferenciate games in pub, clan, league) and compute statistics.
  • The computed score is only here to give an abstract of kill/death/creepKill habilities. It of course does not discuss of the potential role of a heros in a team process.
  • This work should not be considered as an expert guide for each hero. It does not tell what is the best item for each hero. But it does not give absurd recommendation, such as Aghanim for Abaddon or Dagger for Akasha. It only could be used by beginner players.


Statistics confirm knowledges that are related to the game expertise and enable an in-depth analysis of its mechanisms. Parsing a big quantity of game results gives an interesting alternative to the use of expert knowledge, because every player is himself an expert and the emergence of regularities in its behavior can lead to strategical recommendations.

This study is of course partial, but shows that an automatic analysis of DotA games can provide useful knowledge. Just keep in mind that the computer that produced these results does not know DotA, but still can apprehend its behaviour and give useful recommendations.

« June 2011 »
Su Mo Tu We Th Fr Sa
1 2 3 4
5 6 7 8 9 10 11
12 13 14 15 16 17 18
19 20 21 22 23 24 25
26 27 28 29 30

Powered by Plone, the Open Source Content Management System

This site conforms to the following standards: