Atlantis Highlights in Computer Sciences, volume 4

Proceedings of the 3rd International Conference on Integrated Intelligent Computing

Communication & Security (ICIIC 2021)

Prediction of IPL Match Outcome Using Machine

Learning Techniques
Srikantaiah K C1,*, Aryan Khetan1, Baibhav Kumar1, Divy Tolani1, Harshal Patel1
1
Department of CSE, SJB Institute of Technology, Affilated to Visveswaraya Technological University BGS Health &
Education City, Bengaluru-560060, Karnataka, India
*
Corresponding author. Email: [email protected],

ABSTRACT
India's most popular sport is cricket and is played across all over the nation in different formats like T20, ODI, and Test.
The Indian Premier League (IPL) is a national cricketmatch where players are drawn from regional teams of India,
National Team and also from international team. Many factors like live streaming, radio, TV broadcast made this league
as popular among cricket fans. The prediction of the outcome of the IPL matches is very important for online traders
and sponsors. We can predict the match between two teams based on various factors like team composition, batting and
bowling averages of each player in the team, and the team's success in their previous matches, in addition to traditional
factors such as toss, venue, and day-night, the probability of winning by batting first at a specified match venue against
a specific team. In this paper, we have proposed a model for predicting outcome of the IPL matches using Machine
learning Algorithms namely SVM, Random Forest Classifier (RFC), Logistic Regression and K-Nearest Neighbor.
Experimental results showed that the Random Forest algorithm outperforms other algorithms with an accuracy of
88.10%.

Keywords: Cricket, Indian Premier League, Logistic Regression, Machine Learning, Prediction of match
outcome, Random Forest Classifier.

1. INTRODUCTION and others and the entire IPL is controlled by Board of

Control for Cricket in India (BCCI). For the current year
Cricket is an outdoor game which is played by bat and (2021) there are total of 8 Teams in IPL namely, Royal
bowl which includes 2 teams of 11 players each. Cricket is Challengers Bangalore (RCB), Rajasthan Royals (RR),
a teamwork game and is played mostly in three formats and Chennai Super Kings (CSK), Mumbai Indians (MI),
occupies the 2 spots in the list of the most popular sport Kolkata Knight Riders (KKR), Delhi Capitals (DC),
around the World. Like in any sport, there are many factors Punjab Kings (PK) and SunRisers Hyderabad (SRH). The
that plays an importantrole in deciding the winner of the motivation behind this paper includes the answers to
match. Selection of a team is based on the player following questions: “What is the probability of winning
performance and other considerations like pitch factor, the game at a particular venue based on decision to
team size, venue etc. There are many variables and field/bat first on winning the toss?”, “Most dismissals by
constraints which makes The Analysis of Cricket Match a bowler in a match?”, “Does Home Ground have any
Difficult. There are three different formats of Cricket effect on the result of the game?
namely - Tests, Twenty-twenty (T20) and One Day
International (ODI). Cricket is not only a nation game but In this paper We are trying to find out the match winner
also an international game. In this game, every ball is of an IPL match based on the stadium they are choosing
crucial because every ball can change the whole match in and the toss decision using machine learning techniques
Cricket [15, 16]. like SVM, Random Forest, Logistic Regression etc.
Remainder of the paper is organized as follows: The
Indian Premier League (IPL) is a national cricketmatch section 2 is the literature survey, section 3 deals with the
where players are drawn from regional teams of India, problem definition and the architecture. Section 4 deals
National Team and also from international team. It is based with the experimental results. Section 5 talks about the
on 20-20 format and is owned by Celebrities, Businessmen conclusion.

Copyright © 2021 The Authors. Published by Atlantis Press International B.V.

This is an open access article distributed under the CC BY-NC 4.0 license -http://creativecommons.org/licenses/by-nc/4.0/. 399
 Atlantis Highlights in Computer Sciences, volume 4

2. LITERATURE SURVEY machine learning techniques. The model gave an

accuracy of 76%, 67%, and 96% for batsmen, bowlers,
Ahmad et al. [1], predicted the emerging players and all-rounders respectively by using Random Forest
from batsman as well as from the bowlers using machine Algorithm. This model helped them to select the best
learning techniques. Song et al. [2] predicted estimation players of the game and predict outcomes of the match
of the location of a moving ball based on the value of the [25-29].
cricket sensor network. Roy et al. [3] predicted ranking
system which is based on the social network factors and 3. PROBLEM DEFINITION AND
their evaluation in the form of composite distributed ARCHITECTURE
framework using Hadoop framework and MapReduce
programming model is used for processing the data. 3.1. Problem Definition
Priyanka et al. [4], predicted the outcome of IPL-2020
based on the 2008-2019 IPL datasets using Data Mining Given IPL datasets of past 9 years, the main objective
Algorithms with an accuracy of 82.73%. of this paper is to predict the outcome of an IPL match
between two teams based on the analysis of previously
Kansal et al. [5], predicted player evaluation in IPL stored data using Machine Learning algorithms. The
based on the 2008-2019 datasets using Data Mining information will be analyzed and preprocessed. After
Technique. Data mining algorithms are used which gives preprocessing the data will be used to train different
evaluation using player statistics assessing a player's models in order to give the outcomes. We will analyze
performance and determining his base price. They the various datasets and use key variables such as strike
predicted about how to select a player in the IPL, based rate, bowler economy, etc. and feed it as input to an
on every player’s performance history using algorithms algorithm will help us get the probable outcome of a
like decision tree, Naïve Bayes and Multilayer perceptron match [16-19].
(MLP). MLP outperforms better than other algorithms.
Agrawal et al. [6], used Support Vector Machine (SVM), 3.2. Architecture
CTree, and Naïve Baiyes classifiers with accuracies of
95.96%, 97.97% and 98.98% respectively, to predict the The figure 1 represents the architecture of the model
probability of the winner of the matches. Barot et al. [7], which includes different components like datasets, split
predicted the match outcome based on the toss and venue. data, Training, Testing, Supervised Learning models and
Kaluarachchi et al. [8], predicted match outcome Result.
using home ground, time of the match, match type,
winning the toss and then batting first by using Naïve
Bayes classifier. Passi et al. [9], predicted the
performance of players based on the runs and the number
of wickets. Both the type of problems is treated as
classification problems where the list of runs, and list of
wickets are classified in different ranges based on
machine learning algorithms. The Random Forest
algorithm outperforms better than other algorithms. Nigel
Rodrigues et al. [10], predicted the value of the traits of
the batsmen and the bowlers in the current match. This
would help in selecting the players for the upcoming
matches by using past performances of a player against a
specific opposition team by using Multiple Random
Forest Regression.
Wright [11], predicted the possible fixture for a
cricket match based on the various venue, teams, number
of holidays between each match in a fair and efficient Figure 1 Architecture of the model
manner. A metaheuristic procedure is used to progress
from the basic solution to a complex final solution by a
3.2.1. Dataset
technique, Subcost-Guided Simulated Annealing
(SGSA). Maduranga et al. [12], predicted the outcome The first step in the Architecture of model is to collect
of any cricket match by using data mining algorithms and datasets from various sources. The data which is fed into
provided solutions for the approach used by other the model decides how the model acts and reacts. If the
authors. Shetty et al. [13], predicted the capabilities of data is accurate and up-to-date, then we will have
each player depending on various factors like the ground, accurate outcomes or predictions. So, we have collected
pitch type, opposition team and several others by using 6 datasets from Kaggle.com which are as follows [20-21].

400
 Atlantis Highlights in Computer Sciences, volume 4

3.2.1.1. Team wise home and away dataset the match was
played.
The Teamwise Home and Away dataset contains 6 Team 1 Name of team who
columns for the datasets which are as follows: played match.
home_wins, away_wins, home_matches, away_matches, Team 2 Name of team which
home_win_percentage and away_win_percentage. It played match.
describes about the team performance in the home and Toss winner Team that won the
away conditions with their win percentage. The table 1 toss.
shows the dataset and its description [22]. Toss decision Decision of batting
or fielding after
Table 1. Teamwise Home and Away dataset description winning toss.
Column name Column Result Outcome of match
description whether normal or
tie.
Home_wins Tells the number
of matches won by Dl_applied Information on
a team in their whether DL method
home ground. was applied or not.
Tells the number Winner Team that won the
away_wins
of matches won by match.
a team other than Win by runs The number of runs
their home ground. by which the team
Tells the number won.
home_matches
of matches played Win by wickets The number of
by a team in their wickets by which the
home ground. team won.
Away_matches Tells the number Player of match Name of player who
of matches played was awarded player
by a team other of the match.
than their home Venue Stadium in which
ground. match took place.
home_win_percentage Percentage of Umpire 1 On field umpire
matches won when name.
played in home Umpire 2 On field umpire
ground. name.
Away_win_percentage Percentage of
matches won when 3.2.1.3. Player’s dataset
played in ground
other than their The Player’s dataset contains 5 columns namely
home ground. Player_Name, DOB, Batting_Hand, Bowling_Skill and
Country. This dataset tells about the player and his
3.2.1.2. Matches dataset bowling and batting style. The table 3 shows the dataset
column and its description.
The Matches datasets contains 16 columns i.e season,
Table 3. Player’s dataset description
city, date, team1, team2, toss_winer, toss_decision,
result, dl_applied, winer, win_by_runs, win_by_matches Column name Column description
and player_of_the_match, venue, umpire1 and umpire2. Player Name Name of the players
This dataset tells about the matches that are played
between two teams and who was the winner of the match. DOB Date of birth of the players
It also tells about the toss decision taken in the match. Batting Hand Tells whether the players are left-handed
The table 2 shows the dataset column and its description or right-handed.
[23].
Bowling skill Bowling style of players
Table 2. Matches dataset description
Country Name of countries to which the player
Column name Column description belongs.
Season This column tells the
season in which the
match was played.
City The city in which the
match was played.
Date The date on which

401
 Atlantis Highlights in Computer Sciences, volume 4

3.2.1.4. Teams Datasets player_dismissed If the player was given out or not

The teams’ datasets contain a single column named as dismissal_kind What kind of dismissal it was
team1 which shows the various IPL teams. The table 4
shows the dataset column and its description. Fielder Player who caused dismissal

Table 4. Teams dataset description

Column Name Column Description 3.2.1.6. Most_runs_average_strikerate dataset
Team Names of teams participating
in the IPL tournament. The Most_runs_average_strikerate dataset contains 6
columns namely batsman, total_runs, out, numberofballs,
3.2.1.5. Deliveries dataset average, strikerate. This tells about player batting
statistics. The table 6 shows the dataset column and its
The Deliveries dataset contains 20 columns i.e description.
innings, batting_team, bowling_team, over, Ball, Table 6. Most_runs_average_strikerate dataset
Batsmen, Non_striker, Bowler, is_super_over, wide run, description
bye_run, Legbye_run, no_ball_run, penalty_run,
batsmen_run, extra_runs, total_runs, players_dismissed, Column name Column description
dismissal_kind, fielder. The table 5 shows the dataset
Batsman Name of the batsman.
column and its description [24].
Table 5. Deliveries dataset description Total_runs Total runs scored by the
batsman
Column Name Column Description
Out Tells about the number of
Inning Tells us the inning being played times the batsman got out in
the entire IPL career.
Batting_team Tells us name of then batting
team No_of_balls Number of balls faced by the
batsman overall.
Bowling_team Tells name of bowling team
Average The average runs of the
Over Tells the over number being batsman
bowled
Strikerate Tells us about the strike rate
of the batsman.
Ball Tells the ball number of the over

Batsmen Name of batsmen on strike

3.2.2. Split Data
Non-striker Name of batsmen on runner end
In this step the dataset is splitted into two groups, one
Bowler Name of bowler bowling for testing and one for training. The training data is used
to train the Machine learning algorithms using supervised
Is_super_over Tells if over is super over or not learning techniques. The trained model is then tested
using the algorithms and the result is predicted. The
wide_run If there are runs given for wide testing data and training data are divided in the ratio
ball 30:70.
bye_run If there are any bye runs given
3.2.3. Training the model
Legbye_run If there are any leggy runs given
Training is the most important stage in Machine
no_ball_run If the ball was a no ball Learning. In this step the model is trained using training
data to find patterns and make predictions. It results in the
penalty_run Penalty runs due to any reason model learning from the dataset so that it can accomplish
the given task.
batsmen_runs Runs hit by batsmen on the ball
3.2.4. Testing the model
extra_runs Total extra runs given
After training the model, the performance of the
total_runs Total runs in the ball
model is checked. This is done by testing the model with
previously unseen datasets. The unseen Datasets is called

402
 Atlantis Highlights in Computer Sciences, volume 4

the testing datasets. So, in this step the model is tested by

providing the unseen testing data.

3.2.5. Supervised Learning

There are different types of Supervised Machine
Learning Techniques like Logistic Regression, Support
Vector Machine, K-Nearest Neighbor, Random Forest
classifier, linear Regression etc.

4. EXPERIMENTAL RESULTS

4.1. Experimental Setup

The Experiments were carried out on minimum Figure 3 Team winning percentage while bowling first
Hardware requirements that include Processor of Intel
Atom or Intel core i3, Disk space should be at least 1GB, Away win percentage is also being displayed in figure
Operating System must be windows 7 or later, macOS 4 below. From the figure 4, it can be seen that GL has the
and Linux. The machine learning algorithm was highest away win percentage.
implemented using Python 3.8 and Jupyter Notebook.
The Datasets were downloaded from Kaggle.

4.2. Analysis of IPL datasets

The IPL dataset was analyzed, and some features
were extracted out such as team winning percentage
while batting first and team winning percentage while
bowling first which is shown in figure 2 and figure 3.

Figure 4 Teamwise winning percentage in away

matches

Home win percentage is also being displayed in fig 5

below. From the fig 5, it can be seen that RPS has the
highest home win percentage.

Figure 2 Team winning percentage while batting first

Figure 5 Teamwise winning percentage in home

matches

403
 Atlantis Highlights in Computer Sciences, volume 4

The fig 6 gives the information about how many times

a particular team has won the toss and a pie chart is made
from the dataset information depicting the winning
percentage of both batting first and batting second at a
particular venue after winning the toss in fig 7.

Figure 9 Top run scorers

4.3. Performance Evaluation

1) Accuracy: The fraction of correct forecast
in all predictions is known as accuracy. In this
experiment, random forest classifier outruns all the
algorithms by predicting the result with highest accuracy
Figure 6 how many times a particular team won the toss
of 88.10%. The figure 10 shows the accuracies of the
various algorithms implemented.

Figure 10 Accuracy of various algorithms

Table 7 shows about the various algorithms and their

accuracies obtained. It is clear from the table that the
Figure 7 Winning percentage after winning the toss random forest classifier performed better than other
algorithm.
The comparison between the two particular teams
playing the match is shown through graph which could Table 7. Accuracy achieved by the algorithms
help in predicting the winner. Comparison between MI Algorithm Accuracy
and CSK over various seasons is shown over here 88.10%
Random Forest
through figure 8. 49.34%
K-Nearest Neighbor
Logistic Regression 51.40%
SVM 32.6%
Linear Regression 44.05%

Furthermore, 2-fold, 5-fold, 10-fold cross validation

for rfc is also implemented for having better insights in
table 8

Figure 8 Team Comparison

Figure 9 displays the highest run scorers of IPL. From

the fig, it can be seen that Virat Kohli is the highest run
scorer with 5434 runs.

404
 Atlantis Highlights in Computer Sciences, volume 4

Table 8. Cross validation Technique Cricket Sensor Network, IEEE Transactions on

Industrial Electronics, Volume: 59, Issue: 11,
November 2012.
[3] Sarbani Roy, Paramita Dey and Debajyoti Kundu,
Social Network Analysis of Cricket Community
Using a Composite Distributed Framework: From
Implementation Viewpoint, IEEE Transactions
on Computational Social Systems, Volume: 5,
Issue: 1, PP. 64-81, March 2018.
[4] Priyanka S, Vysali K, K B PriyaIyer, Score
Prediction of Indian Premier League- IPL 2020
using Data Mining Algorithms, International
Journal for Research in Applied Science &
Engineering Technology (IJRASET), Volume 8,
Issue II, PP. 790-795.
5. CONCLUSIONS [5] Prince Kansal, Pankaj Kumar, Himanshu Arya,
Aditya Methaila, Player valuation in Indian
Predicting a winner in a sport such as cricket is
premier league auction using data mining
especially challenging and involves very complex technique, International Conference on
processes. But with the introduction of machine learning, Contemporary Computing and Informatics (IC3I),
this can be made much easier and simpler. In this paper, 27-29 Nov 2014
various factors have been identified that contribute to the [6] Shilpi Agrawal, Suraj Pal Singh, Jayash Kumar
results of the Indian Premier League matches. Factors Sharma, predicting results of IPL T-20 Match
that have a major impact on the outcome of an IPL match using Machine Learning, 2018 8th International
include the teams playing, the venue, the city, the toss Conference on Communication Systems and
winner and the toss decision. We have analyzed IPL data Network Technologies (CSNT), 24-26 Nov. 2018.
sets and predicted game results based on player [7] Harshit Barot, Arya Kothari, Pramod Bide,
performance. The methods used in the work to obtain the Bhavya Ahir, Romit Kankaria, Analysis and
final test are Logistic regression, Support Vector Prediction of Indian Premier League, 2020
Machine (SVM), Decision tree, Random Forest classifier International Conference for Emerging
and K-nearest neighborhood. Random Forest Technology (INCET), 5-7 June 2020.
classification (RFC) outperforms the other algorithm. [8] Amal Kaluarachchi, S. Varde Aparna, CricAI:
A classification-based tool to predict the
As for the scope of the future, the focus can be on each outcome in ODI cricket, 2010 Fifth
player’s performance and evaluate that on a regular basis International Conference on Information and
for the season. His ratings for bowling and batting can Automation for Sustainability, 17-19 Dec.
also be predicted. There can be a chance to predict the 2010.
man of the match for the two teams. [9] Kalpdrum Passi and Nirav Kumar Pandey,
predicting player’s performance in one day
ACKNOWLEDGMENTS international cricket matches using machine
We would like to express our sincere thanks to the learning, Volume 8, Computer Science &
Department of CSE, SJBIT for giving us a chance to Information Technology, 2018.
work on this paper and our institution, SJB Institute of
[10] Nigel Rodrigues, Nelson Sequeira, Stephen
Technology for providing immense support throughout
Rodrigues, Varsha Shrivastava, Cricket Squad
the paper. We would also like to extend our thanks to
Analysis Using Multiple Random Forest
Visvesvaraya Technological University for providing us Regression, IEEE Xplore, 1st International
the opportunity to work on this paper. Conference on Advances in Information
Technology, 2019.
[11] M. B. Wright, Scheduling fixtures for New
REFERENCES Zealand Cricket, IMA Journal of Management
Mathematics 16, PP. 99−112, 2005.
[1] Haseeb Ahmad, Ali Daud, Licheng Wang, Haibo [12] Manuka Maduranga, Hatharasinghe,
Hong, Hussain Dawood and Yixian Yang, Guhanathan Poravi, Data Mining and Machine
Prediction of Rising Stars in the Game of Cricket, Learning in Cricket Match Outcome Prediction:
IEEE Access, Volume 5, PP. 4104 – 4124, 14 Missing Links, 5th International Conference for
March 2017. Convergence in Technology (I2CT), Mar 29-
[2] Haryong Song, Vladimir Shin and Moongu Jeon, 31, 2019.
Mobile Node Localization Using Fusion [13] Monali Shetty, Sankalp Rane, Chaitanya
Prediction-Based Interacting Multiple Model in Pandita, Suyash Salvi, Machine learning-based

405
 Atlantis Highlights in Computer Sciences, volume 4

Selection of Optimal sports Team based on the sensing images." Remote Sensing 12, no. 24
Players Performance, Proceedings of the Fifth (2020): 4135.
International Conference on Communication [23] Puttamadappa, C., and B. D. Parameshachari.
and Electronics Systems (ICCES 2020), IEEE "Demand side management of small scale loads
Conference Record, ISBN: 978-1-7281-5371-1, in a smart grid using glow-worm swarm
2020. optimization technique." Microprocessors and
[14] Balasundaram A, Ashokkkumar S, Jayashree D, Microsystems 71 (2019): 102886.
Magesh Kumar S, Data Mining based [24] K. Yu, L. Tan, X. Shang, J. Huang, G. Srivastava
Classification of Players in Game of Cricket, and P. Chatterjee, “Efficient and Privacy-
proceedings of the International Conference on Preserving Medical Research Support Platform
Smart Electronics and Communication Against COVID-19: A Blockchain-Based
(ICOSEC 2020), IEEE Xplore Part Number: Approach”, IEEE Consumer Electronics
CFP20V90-ART; ISBN: 978-1-7281-5461-9. Magazine, doi: 10.1109/MCE.2020.3035520.
[15] Jalaz Kumar, Rajeev Kumar, Pushpender [25] K. Yu, M. Arifuzzaman, Z. Wen, D. Zhang and T.
Kumar, Outcome Prediction of ODI Cricket Sato, “A Key Management Scheme for Secure
Matches Using Decision Trees and MLP Communications of Information Centric
Networks, 2018 First International Conference Advanced Metering Infrastructure in Smart
on Secure Cyber Computing and Grid,” IEEE Transactions on Instrumentation and
Communication (ICSCCC). Measurement, vol. 64, no. 8, pp. 2072-2085,
[16] Prabu, S., Balamurugan Velan, F. V. Jayasudha, August 2015.
P. Visu, and K. Janarthanan. "Mobile https://ieeexplore.ieee.org/document/7138617
technologies for contact tracing and prevention [26] K. Yu, L. Lin, M. Alazab, L. Tan, B. Gu, “Deep
of COVID-19 positive cases: a cross-sectional Learning-Based Traffic Safety Solution for a
study." International Journal of Pervasive Mixture of Autonomous and Manual Vehicles in
Computing and Communications (2020). a 5G-Enabled Intelligent Transportation System”,
[17] Subramani, Prabu, K. Srinivas, R. Sujatha, and B. IEEE Transactions on Intelligent Transportation
D. Parameshachari. "Prediction of muscular Systems, doi: 10.1109/TITS.2020.3042504.
paralysis disease based on hybrid feature [27] K. Yu, L. Tan, M. Aloqaily, H. Yang, and Y.
extraction with machine learning technique for Jararweh, “Blockchain-Enhanced Data Sharing
COVID-19 and post-COVID-19 with Traceable and Direct Revocation in IIoT”,
patients." Personal and Ubiquitous IEEE Transactions on Industrial Informatics, doi:
Computing (2021): 1-14. 10.1109/TII.2021.3049141.
[18] Do, Dinh-Thuan, Tu Anh Le, Tu N. Nguyen, [28] Z．Guo, K. Yu, Y. Li, G. Srivastava, and J. C. -
Xingwang Li, and Khaled M. Rabie. "Joint
W. Lin, “Deep Learning-Embedded Social
impacts of imperfect CSI and imperfect SIC in
Internet of Things for Ambiguity-Aware Social
cognitive radio-assisted NOMA-V2X
Recommendations”, IEEE Transactions on
communications." IEEE Access 8 (2020):
Network Science and Engineering, doi:
128629-128645.
10.1109/TNSE.2021.3049262.
[19] Le, Ngoc Tuyen, Jing-Wein Wang, Duc Huy Le,
Chih-Chiang Wang, and Tu N. Nguyen. [29] Z. Guo, A. K. Bashir, K. Yu, J. C. Lin, Y. Shen,
"Fingerprint enhancement based on tensor of “Graph Embedding-based Intelligent Industrial
wavelet subbands for classification." IEEE Decision for Complex Sewage Treatment
Access 8 (2020): 6602-6615. Processes”, International Journal of Intelligent
[20] Nguyen, Tu N., Bing-Hong Liu, Nam P. Nguyen, Systems，2021, doi: 10.1002/int.22540.
and Jung-Te Chou. "Cyber security of smart grid:
attacks and defenses." In ICC 2020-2020 IEEE
International Conference on Communications
(ICC), pp. 1-6. IEEE, 2020.
[21] Kumar, M. Keerthi, B. D. Parameshachari, S.
Prabu, and Silvia liberata Ullo. "Comparative
Analysis to Identify Efficient Technique for
Interfacing BCI System." In IOP Conference
Series: Materials Science and Engineering, vol.
925, no. 1, p. 012062. IOP Publishing, 2020
[22] Rajendran, Ganesh B., Uma M. Kumarasamy,
Chiara Zarro, Parameshachari B. Divakarachari,
and Silvia L. Ullo. "Land-use and land-cover
classification using a human group-based particle
swarm optimization algorithm with an LSTM
Classifier on hybrid pre-processing remote-

406