Abstract

Boxing effectiveness is strongly influenced by neuromuscular power generation and the efficient transmission of force through the lower extremities; however, traditional evaluation methods tend to assess these variables in isolation across athletes. A biomathematical clustering strategy provides a structured approach to extracting coherent performance patterns from complex, multidimensional biomechanical datasets. The present investigation proposes a biomathematical clustering model to categorise amateur boxers and quantify between-cluster variation, thereby facilitating tailored training interventions. A cohort of 30 amateur competitors underwent a series of standardised biomechanical tests, including Muscle Power (MP), reaction time (RT), rear-leg ground reaction force (GRF) relative to body mass, and maximal cross-punch (MCP) force output. Before analysis, all variables were rescaled through min–max normalisation. Unsupervised classification was executed via K-means clustering. The quality of clustering was assessed using indices of compactness and separation, specifically the Dunn Index and Davies–Bouldin Index. In contrast, the appropriate cluster count was determined using within-cluster sum of squares (WCSS) interpreted via the elbow method. Statistical procedures, including post hoc testing and effect size computation, were applied to evaluate intergroup differences. Additionally, principal component analysis (PCA) was utilised to project the data into a reduced-dimensional space for clearer visual interpretation of cluster distinctiveness. All computational procedures were implemented in Python. The analysis supported a three-cluster configuration. Cluster 3 (40%) exhibited superior performance characteristics, including elevated MP (7,700 ± 3,500 W), reduced RT (0.18 ± 0.03 s), and greater rear-leg GRF (1.55 ± 0.18 BW) relative to Cluster 2 (p ? 0.002; d = 1.35–2.45). In contrast, Cluster 2 (46.7%) was characterised by diminished MP (4,000 ± 1,400 W) and prolonged RT (0.26 ± 0.07 s), whereas Cluster 1 (13.3%) showed moderate values across all measured variables. The biomathematical clustering framework successfully distinguishes discrete neuromechanical profiles among amateur boxers, thereby enabling cluster-specific training strategies and enhancing individual performance optimisation.