Inner pace: A dynamic exploration and analysis of basketball game pace.

Area of Science:

• Sports Analytics
• Computational Statistics
• Basketball Performance Analysis

Background:

• Understanding basketball game dynamics is crucial for strategic optimization.
• Traditional analysis often overlooks intra-game pace variations.
• Novel segmentation methods can reveal nuanced patterns in game tempo.

Purpose of the Study:

• To investigate basketball game pace dynamics and their influence on game outcomes.
• To develop and apply a novel intra-game segmentation approach using K-means clustering.
• To analyze the predictive value of different pace segments using machine learning.

Main Methods:

• Employed K-means clustering on possession duration to segment NBA games (2019-2020 season) into high-frequency (HFS), low-frequency (LFS), and normal-frequency segments (NFS).
• Utilized a sliding window method for segment identification and applied Random Forest and LightGBM models for predictive analysis.
• Incorporated SHapley Additive exPlanations (SHAP) for model interpretability and identified key performance indicators (KPIs) within each segment.

Main Results:

• High-frequency segments (HFS) increase towards quarter ends, driven by rapid transitions and urgency.
• Low-frequency segments (LFS) are prevalent in mid-game phases, indicating strategic tempo control.
• Normal-frequency segments (NFS) decrease as the game progresses, with varying KPI importance across segments.

Conclusions:

• The intra-game segmentation approach provides a finer-grained analysis of basketball game pace dynamics than traditional methods.
• Findings offer actionable insights for optimizing coaching strategies based on real-time game pace.
• Establishes a robust framework for integrating machine learning in sports analysis.