Predict match outcome for second inning of a T20 match. I took past T20 international matches dataset from CRICSHEET and trained a deep neural network to give winning probability of teams after each ball of the game.
Dataset was collected from the CRICSHEET webiste. Cricsheet provides access to ball by ball data for men and women Test matches, ODI, T20 internationals and all IPL seasons. The data is provided in YAML format. I took data for men T20 international matches, parsed it and generated a csv file with required attributes. We can take ODI dataset and run the same model with some changes.
For each ball in a match a feature of ['team-batting', 'team-bowling', 'wicket-left', 'runs-scored', 'target-score', 'balls-bowled', 'batsman-score', 'nonstriker-score', 'batsman-balls-faced', 'nonstriker-balls-faced'] is created. And feature of a match will be a collection of the feature corresponding to each ball in the match. Match's feature is fed to model for training and prediction.
An LSTM neural network model is used for training.
England(Batting) vs South Africa(Bowling): 2020-12-01 | Winner: England
New Zealand(Batting) vs India(Bowling): 2020-02-02 | Winner: India
