Hybrid Ensemble Machine Learning for Constrained Team Selection in Regional Cricket

Abstract

Cricket team selection at the regional level has historically relied on subjective expert judgment rather than data-driven processes. This paper proposes a hybrid machine learning approach that integrates ensemble prediction with an expert-derived fitness formula. Building on prior work using recursive feature elimination and genetic algorithms for fantasy cricket, we adapt the methodology to a single fielded roster with exact role counts and incorporate contextual factors absent from fantasy-sport formulations: pitch suitability, opposition impact, fatigue, and home advantage. A voting ensemble of Random Forest, Gradient Boosting and Ridge Regression is combined with the expert formula through a 60:40 weighted score. On a 51-player dataset the method produces an 11-player team meeting every positional constraint deterministically, in time that scales as n log n. A simulated 300-player pool confirms that the method scales. The approach is implemented as an interactive desktop application for coaches.