M.S. Applied Data Science - Capstone Chronicles 2025

19

Table 7 Model Performance Comparison: Random Forest , NNet, XGBoost and Ensemble by Position R 2 , MAE

R ² = coefficient of determination; MAE = mean absolute error. Higher R ² values indicate better model fit, while lower MAE values indicate better prediction accuracy. Future Performance Forecasting : The 4-week ahead prediction system analyzed 30 active players using time series forecasting models trained on 16 weeks of historical performance data (see Figure 19). Key predictions include Kylian Mbappé's projected performance trajectory showing consistent upward momentum (+1.8 points weekly), while Vinícius Jr. demonstrates stable high-level performance with minor fluctuations (±0.5 points). The forecasting system successfully identified Éder Militão's defensive performance improvement (+2.12 points) These predictions enable proactive tactical adjustments and informed decision-making for upcoming fixtures. Figure 19 Real Madrid Player Performance Forecasting: 16 Week Historical Analysis With 4-Week Predictions

Position/model Forward Random forest

R ²

MAE

0.890 0.980 0.996 0.923 0.500 0.990 0.954 0.777 0.990 0.990 0.969 0.846 0.900 0.920 0.994 0.395 0.821 0.975 0.978 0.735

1.100 0.490 0.229 1.494 3.300 0.390 0.841 1.898 0.220 0.220 0.541 1.246 1.040 0.830 0.195 2.186 1.370 0.560 0.452 1.706

XGBoost Ensemble

Neural network Midfield Random forest

XGBoost Ensemble

Neural network Defense Random forest

XGBoost Ensemble

Neural network Goalkeeper Random forest

XGBoost Ensemble

Neural =network Average Random forest

XGBoost Ensemble

Neural network

Note . Performance metrics evaluated on test set using position-specific SPPS as target variable. Random forest implemented with 100 estimators and max depth of 10. XGBoost used default hyperparameters with early stopping.

66