M.S. Applied Data Science - Capstone Chronicles 2025

16

Figure 13 SHAP Value Distribution for Forward Position Metrics

Figure 15 SHAP Value Distribution for Defender Position Metrics

Midfielder analysis (see Figure 14) revealed pass completion percentage as the dominant predictive feature (SHAP value: 12.927), substantially exceeding progressive passes (4.798) and tackles (2.802). This finding prompted the upward adjustment of pass completion weight from 2.0 to 2.5, recognizing ball retention as the primary midfield contribution to team success. Figure 14 SHAP Value Distribution for Midfielder Position Metrics

Goalkeeper analysis (see Figure 16) demonstrated progressive distance per 90 minutes as the most influential metric (SHAP value: 78.686), followed by total completion percentage (24.147). These findings validated the distribution-focused weighting scheme without requiring adjustments. Figure 16 SHAP Value Distribution for Goalkeeper Position Metrics

The adjusted SPPS models achieved exceptional predictive performance across all positions (see Table 6), with R ² values ranging from 0.913 (midfield) to 0.993 (goalkeeper). These results demonstrate that the SHAP-informed weight adjustments successfully captured position-specific by using expected xG as dependent variable and how the contributions to team performance are showing in Table 17 Int metric appears as first contributor for defense, creating a

Defender SHAP values (see Figure 15) highlighted interceptions (1.907) and blocks (1.708) as primary defensive contributions, with tackles won (1.580) and defensive third tackles (1.061) showing secondary importance. The prominence of interceptions justified the 67% weight increase, while blocks maintained high importance despite slight weight reduction.

63