Statistical approaches to basketball scoring probability maps: from decision-tree-based methods to spatial techniques

Statistical approaches to basketball scoring probability maps: from decision-tree-based methods to spatial techniques

Scoring probability maps offer a powerful tool for analyzing players’ and teams’ shooting performance on the basketball court, providing valuable insights for developing game strategies and personalized training programs. This work presents a complete workflow for generating scoring probability maps, starting from traditional machine learning methods and progressing to advanced spatial statistical approaches. Initially, we explore decision tree-based machine learning techniques, such as Random Forest and AdaBoost, to estimate scoring probability. These methods, which have already been the subject of published work (Zuccolotto et al. (2021) and Zuccolotto et al. (2023)), provide a baseline for assessing shooting performance through predictive modeling. However, to capture the spatial characteristics of shooting behavior, we shift to spatial statistical methods, which offer a more natural framework for this type of analysis. Specifically, we utilize tools such as lorelograms and variograms to investigate spatial correlations and Indicator Kriging to construct scoring probability maps. These spatial statistical techniques, also leading to published research (Carlesso et al. (2024)), allow us to better model the spatial dynamics of shooting performance. By performing a comparative analysis, we evaluate the strengths and limitations of machine learning methods versus spatial statistical approaches in representing shooting dynamics. The proposed workflow is demonstrated through a structured case study involving all the teams of the Italian Basketball First League. Using a non-public dataset enriched with detailed information—such as data on assisted and uncontested shots—we provide novel insights into shooting patterns and performance at both team and player levels. This contribution highlights the evolution from traditional machine learning models to spatial statistical techniques, offering a comprehensive methodology for creating high-quality scoring probability Maps that enhance the understanding of basketball shooting performance.