Machine Learning Methods in Competitive Swimming Analysis: Paradigm Evolution, Architectural Framework, and Future Prospects

In competitive swimming, victory is decided by hundredths of a second, yet traditional biomechanical analysis often struggles with data complexity and delayed feedback.This review addresses the current research fragmentation by proposing an innovative four-dimensional framework (Data, Feature, Model, and Application layers) to systematically categorize the machine learning (ML) landscape in swimming analysis.We deconstruct the racing process into macro-phases (start, swim-through, turns, and finish) and micro-stroke cycles, highlighting how deep learning architectures like Transformers are replacing manual annotation with automated action segmentation.Furthermore, the study identifies key performance predictors-such as Intra-cyclic Velocity Fluctuations (IVV), Stroke Index (SI), and Countermovement Jump (CMJ) impulse-as critical inputs for robust feature engineering.By integrating Explainable AI (XAI), this study bridges the gap between complex "black box" models and actionable coaching insights.Ultimately, we provide a methodological roadmap for building "digital twins" of athletes, shifting swimming science toward a synergy of biomechanical depth and computational precision.