A SCALABLE BIG DATA ANALYTICS FRAMEWORK FOR ODI CRICKET MATCH OUTCOME PREDICTION USING ENSEMBLE MACHINE LEARNING AND APACHE SPARK

Abstract

The second most-watched sport in the world, with between 2.5 and 3 billion fans worldwide, is undergoing a data revolution through big data analytics and machine learning (ML). This study introduces a scalable predictive model using Random Forest (RF), Decision Tree (DT), and Linear Regression (LR) with Apache Spark ML to predict the scores of the teams from the largest One-Day International (ODI) dataset (2,379 matches with 23 features) from December 2002 to September 2023. The exploratory data analysis revealed that the mean total score was 251.6 runs (SD = 52.4); the highest correlation was for runs_last_5_overs (r = 0.42, p < 0.001), followed by innings wickets (r = −0.24, p < 0.001). Runs_last_5_overs was found to be the most important variable, with an importance of 0.38, which aligns with its significance in an ODI match, given the importance of death-over batting in contemporary cricket. It outperformed Decision Tree RMSE = 25.9 and Linear RMSE = 38.2, and integration with Spark ML reduced computational processing time by around 40% compared to traditional pipelines. The results validate the potential of ensemble learning to be the best paradigm for the big data distributed setting of cricket score prediction and will provide insights for team strategy, player selection, and in-play analytics