Purpose: Use ball-by-ball data from the Indian Premier League cricket tournament and machine learning techniques to predict match outcomes based on events occurring in the first inning of a match.Approach: Twelve predictor variables were entered into machine learning models (forward stepwise logistic regression using Akaike’s Information Criterion (AIC); forward stepwise logistic regression using Bayesian Information Criterion (BIC); random forests; naïve Bayes classifier), with match outcome as the dependent variable. Findings: The AIC model reported the highest accuracy in both the training and test datasets (69.92% and 67.18%, respectively). This model contains total runs scored, winning the coin toss, and home-ground advantage as positive predictors, and number of balls with no runs scored and number of balls with one run scored as negative predictors. All four models found that total runs scored in an inning was the most important predictor of match outcome, and no model included number of wickets lost as a predictor, although there could be an indirect effect through total runs scored. Originality: This study is novel in that it used both pre-match variables (home-ground) advantage and real-time measures (e.g., how many runs were scored in the powerplay) in a machine learning context to classify match results. The results can be used to adapt in-game tactics to maximize advantages of batsmen in favorable contexts.