League of Legends is a MOBA video games, where 2 teams of 5 players compete against each other to destroy the opposing base. Its a team-based strategy game, where the players have over 140 champions to choose from to make plays, secure kills, and destroy enemy towers as you make your way towards the opponent’s base on the Summoner’s Rift. Because it is a team-based strategy game, it is important to pick and choose the correct champions that fit your team composition. Depending on the team compositions, the game length can vary, from passive late game compositions to aggressive early game compositions.
The League of Legends Dataset consists of League of Legends competitive matches between 2015-2017. The matches include the NALCS, EULCS, LCK, LMS, and CBLoL leagues as well as the World Championship and Mid-Season Invitational tournaments. In our project we used the following variables:
Year
- Year the match took place inblueTeamTag
- Blue Team’s namebResult
- Result of the match for Blue Team - 1 is a win, 0 is a lossredTeamTag
- Red Team’s namerResult
- Result of the match for Red Team - 1 is a win, 0 is a lossblueTopChamp
- Name of Blue Team’s champion in the top positionblueMiddleChamp
- Name of Blue Team’s champion in the middle positionblueJungleChamp
- Name of Blue Team’s champion in the jungle positionblueADCChamp
- Name of Blue Team’s champion in the ADC positionblueSupportChamp
- Name of Blue Team’s champion in the support positionredTopChamp
- Name of Red Team’s champion in the top positionredMiddleChamp
- Name of Red Team’s champion in the middle positionredJungleChamp
- Name of Red Team’s champion in the jungle positionredADCChamp
- Name of Red Team’s champion in the ADC positionredSupportChamp
- Name of Red Team’s champion in the support positiongamelength
- Game length in minutesLeague
- League or Tournament the match took place inTeam
- Which team (blue or red) killed the victimx_pos
- x-coordinate for every deathy_pos
- y-coordinate for every deathResearch Questions
Our group was interested in learning how the game changed over time in terms of competitive teams and popular champions. In our first graph we explored Year
, blueTeamTag
, redTeamTag
, bResult
and rResult
variables to see the top 5 teams that won the most games from 2015 to 2017. We had to combine blueTeamTag
with redTeamTag
and bResult
with rResult
so we can have one cohesive data set to analyze from. We decided to make a faceted scatterplot so we can clearly see the trends in top 5 teams over the years.
We can see that SKT has been consistently in the top 5 teams from 2015 to 2017 and they had the most games won from years 2015 to 2017. We can see that especially in 2015 where SKT had more than 90 games won, but all other teams have below 60 wins. Other teams such as kt, TSM, and SSG have been in the top 5 twice. Another trend we noticed was that the number of games won increases yearly with the exception of SKT. The lowest amount of games won in the top 5 teams in 2015 is around 47. In 2016, this number increases to 66 and in 2017 it is at 76. We can infer that the other teams were trying to catch up to SKT and try to improve their skills as SKT had over 90 games won since 2015. The above graph suggests that each year there is a variety of top 5 teams that get the most teams while 1 or 2 teams stay consistently in the top 5. Furthermore, we can conclude a greater trend about this game that each year teams are improving to win more and more games as the game gets more competitive because the amount of games that teams win increases. We can also conclude that there are more games being held in League of Legends probably because the popularity is growing year by year.
In graphs 2, 3, and 4, we explored the Year
, blueTopChamp
, blueJungleChamp
, blueMiddleChamp
, blueADCChamp
, blueSupportChamp
, redTopChamp
, redJungleChamp
, redMiddleChamp
, redADCChamp
, and redSupportChamp
variables to see the trends in popular champtions from 2015 to 2017. We made the variable names uniform and combined the blue and red team champions to analyze the data. Then, we filtered the datasets by year, so we can make wordclouds for them. We chose to use a wordcloud to represent our data as we can indicate the most popular champions by year through the size of the words and see clear groups through the colors. Furthermore, the wordcloud made it easy to plot many champions at a time and clearly be able to read their names.
In 2015, the most popular champions were Thresh, Maokai, Reksai, and Sivir. In 2016, the most popular champions were Elis, Braum, Lucian, and Reksai. In 2017, the most popular champions were Gragas and Varus. With the exception of Reksai, no champion has appeared twice in the wordclouds. We can infer that the most popular champion are the strongest champions as they are picked the most often. It is rare to see champions being the most popular in consecutive years, which suggests that the game developers change the champions’ stats (abilities, powers) when they get too strong as they do not want couple champions to dominate the game. We also noticed that in 2015, Elis is colored in orange and the letters are small, which suggests that she was not a popular champion, but in 2016, she is one of the most popular champions as her name is in grey and her name has increased in size. This also happened to Varus from 2016 to 2017. We can infer that game developers like to increase the power of weaker champion and decrease the power of stronger champions to add variety to the game. In conclusion, from 2015 to 2017 we can see that the most popular champions per year varies, but it follows a trend of less popular champions becoming more popular the year after and most popular champions becoming less popular the year after.
In order to answer this question, we focused on the variables each of the champions picked for both blue team and red team in each roles (blueTopChamp
, blueJungleChamp
, blueMiddleChamp
, blueADCChamp
, blueSupportChamp
, redTopChamp
, redJungleChamp
, redMiddleChamp
, redADCChamp
, redSupportChamp
), the length of the game (gamelength
), and the different regions (League
).
For the champions picked, we combined all ten champion variables to account for every champion picked in competitive games. "
We then found the top 10 most picked champions by regions. We only focused on the Top 4 regions, NALCS, EULCS, LCK and LPL. However, the data is missing LPL, hence we will only focus on the first three regions. From our results, EULCS picked Gragas, Thresh, Reksai, Braum, Sivir, Lucian, Elise, Corki, Gnar, and Alistar as their top 10 most picked champions, respectively, LCK picked Alistar, RekSai, Elise, Maokai, Gragas, Corki, Sivir, Braum, Ashe, and Thresh as their top 10 most picked champions, respectively, and NALCS picked Gragas, RekSai, Braum, Karma, Maokai, Corki, Sivir, Ashe, Lucian, and Elise as their top 10 most picked champions, respectively. The three leagues have many champions in common, such as Gragas, Alistar, RekSai, Braum, Sivir, and Elise appearing in all three regions’ top 10 most picked champions. Thresh only appeared in the Top 10 most picked champions for EULCS and LCK. Lucian only appeared in the Top 10 most picked champions for EULCS and NALCS. Ashe and Maokai only appeared in the Top 10 most picked champions for LCK and NALCS. Karma only appeared in the Top 10 most picked champions for NALCS. Gnar only appeared in the Top 10 most picked champions for EULCS.
One thing to note before taking the graph into consideration is the fact that there are more LCK games recorded than there are of other leagues.
To better see the differences in picked champions, we will make a faceted bar graph of the most commonly picked champions, which are Gragas, Thresh, RekSai, Braum, Elise, Alistar, Maokai, Corki, Lucian and Sivir."
From our faceted bar graph of Top 10 champions by the top 3 regions, we can better examine the difference between pick rates of the Top 10 most popular champions. On the x-axis, the Top 10 champions are shown. On the y-axis, the number of times the champion was picked is shown. From the graph, we can determine that Gragas is most picked for EULCS and NALCS, while Alistar is most picked for NALCS.
The overall distribution from the graph makes sense, as champions such as Alistar, Elise, Maokai, Sivir and Thresh are relatively old champions. It also makes sense that Lucian, Corki and Gragas are on the list because these three champions have been used in multiple roles.
Since we know the Top 3 leagues share the Top 10 most picked champions, we will compare gamelength by all regions to determine whether each regions vary in gamelength.
In the histogram, the x-axis represents game length of each games, and the y-axis represents the number of games. For all regions, we can examine that the graph is uni-modal and right-skewed, with the peak around 30 to 40 minutes. This is more evident in the Top 3 regions, where more games are played and recorded. Most games last between 20 minutes and 60 minutes, with few outliers of games lasting beyond 60 minutes in EULCS, LCK, LJL, LMS, NALCS, TCL and at the Worlds Championship.
We wanted to explore statistical insights into the game and additionally what factors affected the outcomes of games. With our data available we wanted to see if any information could help inform players of the game on things to watch out for when playing themselves. We first started by plotting all of the kills recorded within our data onto a heat map given the kills x and y coordinates and which team did the killing. We chose a hexagonal heat map as it was a good balance between a continuous one and a tile one given the nature of our map containing solid terrain with varied edges.
From our graph we can clearly see increased areas of kill activity going down the 3 lanes (top, mid, bot) especially the mid lane, as well as within the river (diagonally top left to bottom right) for both kills made by teams starting on the blue side and red side. Clear hexagons indicate no kills were recorded there, and from a image of the map one can see that it is because of terrain occupying that area. We also see the highest gradient hexagon for blue side kills within the Baron pit, a semi-circle within which a neutral boss can be fought, the killer of which receives a big gold boost and temporary buff to all their teammates. Knowing that these areas have a high kill amount, players should be wary when in these areas especially if they are playing on the red side. It seems that the red side has a unique disadvantage given that the walls of the objective only face them. Further investigating this disparity between sides we decided to run a binomial test against the win-rates given side across all competitive games. Players should be wary of this statistical difference in sides, but as the sides are decided randomly they cannot do anything about it.
##
## Exact binomial test
##
## data: wins
## number of successes = 4146, number of trials = 7620, p-value =
## 1.452e-14
## alternative hypothesis: true probability of success is not equal to 0.5
## 95 percent confidence interval:
## 0.5328311 0.5553241
## sample estimates:
## probability of success
## 0.5440945
With a p-value of close to 0, we reject the null that the side one team gets is independent from their outcome, noting therefore there is some relationship between side and winning. We see from our code that those teams that are on the blue side had a 54.40945% win rate, thus our previous investigation into the map shape and kill locations may have played a role in deciding the outcome of games given that the difference in sides is statistically significant.
Through our research questions and investigation into this League of Legends dataset we hope to have provided insights into the nature of League of Legend’s competitive scene and the ways in which games are won and lost. We noted the dominance of SKT, as other teams were trying to catch up to and try to improve their skills while SKT had over 90 games won since 2015. We also saw that each year there is a variety of top 5 teams that get the most teams while 1 or 2 teams stay consistently in the top 5. Furthermore, we concluded a greater trend about this game that each year teams are improving to win more and more games as the game gets more competitive because the amount of games that teams win increases. We can also conclude that there are more games being held in League of Legends probably because the popularity is growing year by year. Moreover, we made the inference that the most popular champion are the strongest champions as they are picked the most often. It is rare to see champions being the most popular in consecutive years, which suggests that the game developers change the champions’ stats (abilities, powers) when they get too strong as they do not want a couple champions to dominate the game. We can infer that game developers like to increase the power of weaker champion and decrease the power of stronger champions to add variety to the game. We also noted that between regions the most popular champions differed, indicating that different leagues have different competitive scenes and strategies to win. Additionally we found that most games last between 20 minutes and 60 minutes, with few outliers of games lasting beyond 60 minutes in EULCS, LCK, LJL, LMS, NALCS, TCL and at the Worlds Championship. Finally, newcomers to League of Legends will find of particular interest after the choices made within the competitive scene for champion picks the inherent benefit of playing on the blue side vs. red. We hope that those interested in the competitive scene itself will find the insights into popular teams and their performances across the years potentially useful in correlation to their favorite players and how such player‘s may or may not have contributed to their favorite team’s success over time, and we hope our insights will be beneficial and informative to all who view them.