1. IPL Player Ranking
Group 3
Survesh Chauhan FT153047
Krishna Chaitanya FT153094
Sumit Arora FT153107
Venkataraman M FT152059
Steeve Renold M FT152063
2. IPL Auction Overview
• 8 Seasons has been conducted successfully with 8 Franchises
• IPL Lagged the ranking system due to which would clear picture, where
a player would be easily compared to his counter part based on IPL
performance.
• Below table gives brief overview of money spent country wise
3. Datacollectionandcleansing
• Data Collection:
Data was collected from www.Cricinfo.com , www.iplT20.com and www.thatscricket.com
Data regarding Batting, Bowling and Auction was collected spread over 8 seasons of IPL.
• Data Cleansing:
Excel was used for initial clean up, where Trim, Formatting and consolidation
SAS was used in later cleansing where Proc SQL was used to aggregate the data of 8 seasons,
Merge the data sets.
Batting
Player
Matches
Innings
Runs
Balls Faced
Not Outs
50s
100s
30s
Bowling
Player
Matches
Innings
Runs
Overs
Wickets
5 WHs
4 Whs
Auction
Player
Country
Franchise
Price
4. User variables
• IsFit, a binary Variable was created were cutoff found from
mean was used to detect whether a player is eligible for the
further analysis.
• Logit Score, a numerical variable was created were exponential
value of model was stored and further used to detect the
probability score
• Prob Score, a Numerical Variable was created which was used
to rank the players both for batting and bowling.
Modified Variables
• Batting – Average and Strike rate were modified for the aggregated data
• Bowling – Average, Strike Rate and Economy were modified for the
aggregated data
5. Batting – Logit Model
Prob(scorei) = exp(Bo + B1*Runsi + B2*Batting_Averagei)
1+ exp(Bo + B1*Runsi + B2*Batting_Averagei)
Where: i is player
• Prob(scorei) is predicted probability that player is selected by our
model. We have used this to measure top 18 batsmen from the list
of 430 batsmen that batted in IPL in last 8 seasons.
• B0, B1, B2 are respective Beta weights for batting data for last 8
years
6. Bowling – Logit Model
Prob(scorei) = exp(Bo + B1*Wicketsi + B2*Economyi +
B3*Averagei + B4*Strikeratei)
1+ exp(Bo + B1*Wicketsi + B2*Economyi +
B3*Averagei + B4*Strikeratei)
Where: i is player
• Prob(scorei) is predicted probability that player is selected by
our model. We have used this to measure top 18 bowlers from
the list of 305 bowler that bowled in IPL in last 8 seasons.
• B0, B1, B2, B3, B4 are respective Beta weights for bowling
data for last 8 years
7. Methodology
• Variable ‘isfit’ is created to run LOGIT model, which selects players
with Innings > mean (innings) in the batting/blowing data set.
• data batnew;set batnew;
isfit=inns>17;run;
• Model was run to find Beta weights from dataset of last 8 seasons.
proc logistic data=batnew;
model isfit(event='1')=SR runs AVG;
ods output ParameterEstimates=model_batnew;
run;
• These weights were then used to find probability score for player
batting/bowling in last 3 years (2014, 2013, 2012).
• Rankings were given to players based on probability score i.e. higher
the score higher the ranking.
10. Results - Bowling
• Likelihood Ratio was Significant.
• Bowling Results were significant for Average, Strike rate, Economy
and Wickets
• Average = Negatively correlated
• Strike Rate = Positively correlated
• Economy = Positively Correlated
• Wickets = Positively Correlated
11.
12. Results - Batting
• Likelihood Ratio was significant
• Batting Results were significant for Average and Runs
• Average = Negatively correlated
• Runs = Positively Correlated
13. Limitations and further
analysis
• Regression model was used to find the correlation between
price and performance but there was only one variable (
Wicket in Bowling and Runs in Batting ) so we dropped the
model.
• Man of the match data lacked the details over criteria for the
winning due to which there was ambiguity while apply to the
model so MoM data was not used.
• IPL performance can be processed based on ground, country
and foreign player
14. Conclusion
• The logit equation can be used to detect the ranking of the IPL
player depending on the IPL performance.
• Same model can also be used to rank in BigBash, BPL and
several other popular leagues
• Model can also be used to predict the ranking based on earlier
performance of the new player.
• Ranking gives the clear picture and comparison ground to buy
the players totally based on the performance.