Unsupervised Learning Based Evaluation of Player Performances
- PDF / 1,808,835 Bytes
- 10 Pages / 595.276 x 790.866 pts Page_size
- 4 Downloads / 171 Views
S.I. : ARIIS
Unsupervised Learning Based Evaluation of Player Performances Avijit Bose1 · Sannoy Mitra1
· Souham Ghosh1 · Raima Ghosh1 · Tiyash Patra1 · Satyajit Chakrabarti1
Received: 14 March 2020 / Accepted: 25 September 2020 © Springer-Verlag London Ltd., part of Springer Nature 2020
Abstract In the following paper, we design a model that uses real-time data to segregate players into categories according to their performances in the T-20 tournaments. The data are gathered from reliable websites, cleaned and analysed through cluster space maps based upon certain proposed formula. A thorough research on players’ statistics with different unsupervised clustering algorithms in machine learning and deep learning models is documented and compared through silhouette scores. They are classified based on their strength according to bowlers, batsmen and all-rounders. A comparative study of machine learning algorithms with its deep learning counterparts using auto-encoder is also shown. The paper depicts how the models perform on the given dataset and concludes with the most effective model. Keywords T-20 · Cluster space maps · Unsupervised clustering · Silhouette score · Playing eleven · Auto-encoder
1 Introduction Cricket is one of the most popular games in India. Over the years, shorter formats of the game have gained more popularity. Due to the advent of limited over cricket, team selection on the basis of each player’s strength becomes necessary. Statistically, T-20 cricket has one of its major loyalties in India. Indian premier League, a franchise-based cricket league has had over ten seasons with players from every part of the world taking part in it. With a varied pool of players available, it has become essential for team management to segregate players according to demands.
2 Previous work Cricket is one of the team games played over 50 countries in different levels [1]. This game has manifested itself in many formats. Be it the matches which took days to end or the recent T20 formats that take only half a day, we have come
B
Sannoy Mitra [email protected] Avijit Bose [email protected]
1
Institute of Engineering and Management, Salt Lake Electronics Complex, Y-12 Gurukul, Street Number 18, EP Block, Sector V, Bidhannagar, Kolkata, West Bengal 700091, India
across a long path. The most popular T-20 League is currently the Indian Premier League (IPL) [2,3]. The teams have been formed on the basis of the players bought by bidding. There have been huge franchises who bid for a marquee player so that the team can get the best proportionality of players of every kind. Cricket is highly driven by statistics, and it is one of those games which give a lot of numerical data, that can be analysed [4,5]. There are ample research works that help in predicting the result of the match. The winning probability of a team is designed by using their bowling figures and career average runs which reflect their capability to dismiss the opposition and chase down the target, respectively [6]. A model had been de
Data Loading...
