Vector quantization, density estimation and outlier detection on cricket dataset

Abstract

This study aims to apply unsupervised machine learning algorithms on Cricket players' career statistics dataset. K-means clustering algorithm is used to find the natural grouping that exists within the cricket players using player's batting average, strike rate, bowling average, economy etc. as input features - in this case players are grouped into 3 groups. Further separate probability density models are fitted for batsmen, bowlers and all-rounding players using appropriate player's performance metrics as input features and using these models, outstanding players are identified. Similar method is used to identify match winning players, where the differences between player's performance metrics and team's average performance metrics are used as input features. The results obtained from this study seem to correlate with expert generated results where they used point based system to rank the players. This kind of statistical analysis of sports data plays a vital role in team planning and exploiting opponents' weakness.