Predictive Team Selection Using Player Performance Analytics

Project Overview

The client’s team plays in the world’s third-most-expensive sports league. Most of the club’s expenses are incurred by paying its superstar players’ salaries with no guarantee of on-field success or team camaraderie. Because of the attractiveness of a player’s star power, teams overlooked hundreds of other talented but unknown players with considerably better ROI. Indium Software used Advanced Predictive Analytics and Data Mining to discover and evaluate the players who would be the best investments for the squad from the hundreds of available players in the auction.

About Client

The client is a technology-centric sports consultant that advises professional teams across different sports on strategies that lead to performance enhancement. They excel in collecting extensive firsthand data from each player and game, transforming it into strategic insights for team owners or coaches to act upon.

Business Requirements

The specific business need in question was regarding the world’s 3rd most expensive sporting league, for which each city’s team would need to bid for players in the upcoming player auction. Traditionally, team owners would bid for players based on a combination of the player’s reputation and the coach’s personal opinions. This led to all teams bidding exorbitant sums for a small group of famous players who were in many cases not ideally suited for the teams bidding for them. Additionally, there was no bidding consultant capable of advising on the performance or playing style of each of the hundreds of relatively unknown and overlooked but potentially talented players.

The client has partnered with Indium Software for the following:

  • Recommendations on player bidding: The client seeks analytical reasoning and statistical evidence to advise which players to bid for in the player auction.
  • Ranking of promising players: The client requires a list of the best players based on their positions. These rankings will be developed using Composite Performance Indicators (CPIs) and domain knowledge. The rankings should be objective, comprehensive (considering 50+ criteria), and account for players’ current form.
  • Coaches’ ability to infer player suitability: The rankings should enable coaches to scan and determine which players best fit their team’s needs. The accompanying analytical metrics should provide deep insights into the players’ performance and characteristics.

A) Descriptive Method:

Batting index concepts: The Composite Performance Index for batsmen compares their performance based on the following factors:

Bowling index concepts: The Composite Performance Index for bowlers compares their performance based on the following factors:

  • Strike Rate Index: The number of balls taken to get a wicket.
  • Economic Rate Index: Number of runs conceded per over.
  • Performance in different phases of T20 matches.
  • Comparison of strike rate and economic rate with other players.
  • Performance in recent tournaments.

B) Predictive Method:

Descriptive index creation: Initially, a descriptive index was created to rank the players. The calculations for the descriptive index were refined iteratively until the indexing rank matched the “original rank.” The descriptive index served as the dependent variable.

Training the model: A model was trained to determine player characteristics (independent variables) and their weights for ranking. Variable selection was performed using multiple models such as Linear Regression, Random Forest, and Recursive Feature Elimination technique.

Iterative model refinement: The model was iterated, and transformations were applied to the variables to test its performance. This was necessary as absolute strike rates, economic rates, and batting and bowling averages may have skewed distributions.

Significant features of the Predictive model included:

  • Comparative analysis between players to adjust statistics for game variations and discrepancies.
  • Differential weightages of tournaments to account for varying conditions, players’ improvements, and optimization for the upcoming league.
  • Custom methods to average the accuracy of the CPI predictions.