Predicting How Points
End in Tennis
Simon Pang, Xinxiao Jiang, Lindsay Spencer, Jacob Low, Tan Dun Yong
Introduction
AO is Arguably the largest sporting event in Australia
Point ending type gives insight on player performance
Tracking system follows ball movements and player movements
Aim is to construct a classification model to correctly classify outcomes
Methodology
Random Forest & Bagging
Initial bagging method set a high benchmark (91%)
Transformation to random forests, returned lower score
Added in gendervariable
Increased the amount of variables, no effect on accuracy
SVM
Intuitive method for separating classes
Not appropriate to multiclass classification, led to lower score
XGBoost
Reduces both variance and bias
Cross-validation to find the best model from XGBoost through lowest logarithmic
loss
Increased prediction accuracy on outcome (Unforced and Forced)
Created WINNERvariable to isolate variables that fulfil the winner criteria
A better predictor than other methods used