import warnings
warnings.filterwarnings('ignore')

#Session 1
from statsbombpy import sb
import pandas as pd

comps = sb.competitions()
wsl = comps[comps['competition_name'].str.contains("Women's Super League", case=False, na=False)]
print(wsl[['competition_id', 'season_id', 'season_name']]);

    competition_id  season_id season_name
25              37        281   2023/2024
26              37         90   2020/2021
27              37         42   2019/2020
28              37          4   2018/2019

from statsbombpy import sb
import pandas as pd

matches = sb.matches(competition_id=37, season_id=281)
print(matches.shape)
matches.head();

(132, 55)

print(matches[['match_date', 'home_team', 'away_team', 'home_score', 'away_score']].head(10))

   match_date                   home_team                   away_team  \
0  2023-11-05  Brighton & Hove Albion WFC         Manchester United W   
1  2024-03-23         Manchester City WFC         Manchester United W   
2  2024-04-21         Manchester United W     Tottenham Hotspur Women   
3  2024-04-28               Aston Villa W         West Ham United LFC   
4  2023-12-09         Manchester City WFC               Aston Villa W   
5  2024-04-17                 Chelsea FCW               Aston Villa W   
6  2023-12-17               Aston Villa W  Brighton & Hove Albion WFC   
7  2024-03-16                 Everton LFC               Aston Villa W   
8  2024-02-03               Aston Villa W            Bristol City WFC   
9  2023-10-08               Liverpool WFC               Aston Villa W   

   home_score  away_score  
0           2           2  
1           3           1  
2           2           2  
3           1           1  
4           2           1  
5           3           0  
6           1           0  
7           1           2  
8           2           2  
9           2           0

chelsea_matches = matches[
    (matches['home_team'] == 'Chelsea FCW') | (matches['away_team'] == 'Chelsea FCW')
].copy()

print(chelsea_matches.shape)
chelsea_matches[['match_date', 'home_team', 'away_team', 'home_score', 'away_score']]

(22, 55)

chelsea_matches.to_csv("chelsea_matches_2023_24.csv", index=False)

#Session 2
from statsbombpy import sb
import pandas as pd
import time

all_events = {}

for match_id in chelsea_matches['match_id']:
    events = sb.events(match_id=match_id)
    all_events[match_id] = events
    time.sleep(0.5)  # small pause to be polite to the API

print(f"Pulled events for {len(all_events)} matches");

Pulled events for 22 matches

def summarize_match(match_id, match_row, events):
    is_home = match_row['home_team'] == 'Chelsea FCW'
    chelsea_team_name = 'Chelsea FCW'
    opponent = match_row['away_team'] if is_home else match_row['home_team']
    
    shots = events[events['type'] == 'Shot'].copy()
    
    chelsea_shots = shots[shots['team'] == chelsea_team_name]
    opponent_shots = shots[shots['team'] != chelsea_team_name]
    
    chelsea_xg = chelsea_shots['shot_statsbomb_xg'].sum()
    opponent_xg = opponent_shots['shot_statsbomb_xg'].sum()
    
    chelsea_goals = match_row['home_score'] if is_home else match_row['away_score']
    opponent_goals = match_row['away_score'] if is_home else match_row['home_score']
    
    if chelsea_goals > opponent_goals:
        result = 'W'
    elif chelsea_goals < opponent_goals:
        result = 'L'
    else:
        result = 'D'
    
    return {
        'match_id': match_id,
        'date': match_row['match_date'],
        'opponent': opponent,
        'venue': 'Home' if is_home else 'Away',
        'result': result,
        'goals_for': chelsea_goals,
        'goals_against': opponent_goals,
        'xg_for': round(chelsea_xg, 2),
        'xg_against': round(opponent_xg, 2),
        'xg_diff': round(chelsea_xg - opponent_xg, 2)
    };

summaries = []

for _, row in chelsea_matches.iterrows():
    match_id = row['match_id']
    events = all_events[match_id]
    summary = summarize_match(match_id, row, events)
    summaries.append(summary)

match_level_df = pd.DataFrame(summaries)
print(match_level_df.shape)
match_level_df

(22, 10)

match_level_df.to_csv("chelsea_match_level_2023_24.csv", index=False)

#Session 3
import pandas as pd

match_level_df = pd.read_csv("chelsea_match_level_2023_24.csv")
match_level_df.head()

print(match_level_df['result'].value_counts())
print()
print(match_level_df[['goals_for', 'goals_against', 'xg_for', 'xg_against', 'xg_diff']].describe())

result
W    18
L     3
D     1
Name: count, dtype: int64

       goals_for  goals_against     xg_for  xg_against    xg_diff
count  22.000000      22.000000  22.000000   22.000000  22.000000
mean    3.227273       0.818182   2.543636    0.981818   1.562273
std     1.900672       1.220319   1.172618    0.596215   1.454756
min     0.000000       0.000000   0.890000    0.140000  -1.820000
25%     2.000000       0.000000   1.735000    0.560000   0.540000
50%     3.000000       0.000000   2.450000    0.920000   1.670000
75%     4.000000       1.000000   2.890000    1.355000   2.412500
max     8.000000       4.000000   5.190000    2.710000   4.750000

import matplotlib.pyplot as plt

match_level_df['xg_diff'].hist(bins=15)
plt.xlabel('xG Differential (Chelsea xG - Opponent xG)')
plt.ylabel('Number of Matches')
plt.title('Distribution of Chelsea xG Differential, 2023/24')
plt.show()

match_level_df.boxplot(column='xg_diff', by='result')
plt.title('xG Differential by Match Result')
plt.suptitle('')  # removes the auto-generated subtitle pandas adds
plt.xlabel('Result')
plt.ylabel('xG Differential')
plt.show()

plt.scatter(match_level_df['xg_for'], match_level_df['goals_for'])
plt.plot([0, match_level_df['xg_for'].max()], [0, match_level_df['xg_for'].max()], 'r--', label='Perfect xG conversion')
plt.xlabel('xG For')
plt.ylabel('Actual Goals For')
plt.title('Chelsea: Actual Goals vs. xG per Match')
plt.legend()
plt.show()

# Any matches with 0 xG? Missing data? Extreme outliers?
print(match_level_df[match_level_df['xg_for'] == 0])
print(match_level_df.sort_values('xg_diff').head(3))  # worst xG diff matches
print(match_level_df.sort_values('xg_diff', ascending=False).head(3))  # best xG diff matches

Empty DataFrame
Columns: [match_id, date, opponent, venue, result, goals_for, goals_against, xg_for, xg_against, xg_diff]
Index: []
    match_id        date             opponent venue result  goals_for  \
10   3913105  2023-12-10          Arsenal WFC  Away      L          1   
16   3913158  2024-03-24  West Ham United LFC  Away      W          2   
7    3913138  2024-02-16  Manchester City WFC  Home      L          0   

    goals_against  xg_for  xg_against  xg_diff  
10              4    0.89        2.71    -1.82  
16              0    1.01        1.10    -0.09  
7               1    1.47        1.45     0.02  
    match_id        date                    opponent venue result  goals_for  \
13   3913081  2023-11-04               Aston Villa W  Away      W          6   
12   3913077  2023-10-22  Brighton & Hove Albion WFC  Home      W          4   
3    3913180  2024-05-05            Bristol City WFC  Home      W          8   

    goals_against  xg_for  xg_against  xg_diff  
13              0    5.19        0.43     4.75  
12              2    4.94        1.39     3.55  
3               0    4.56        1.41     3.15

print(match_level_df.sort_values('goals_for', ascending=False).head(5)[['opponent', 'goals_for', 'xg_for', 'xg_diff']])

               opponent  goals_for  xg_for  xg_diff
3      Bristol City WFC          8    4.56     3.15
15  Manchester United W          6    3.30     2.94
13        Aston Villa W          6    5.19     4.75
11   Leicester City WFC          5    2.49     1.24
19        Liverpool WFC          5    2.91     2.15

#Session 4
import pandas as pd
import numpy as np

match_level_df = pd.read_csv("chelsea_match_level_2023_24.csv")
match_level_df.head()

result_map = {'W': 3, 'D': 1, 'L': 0}
match_level_df['result_points'] = match_level_df['result'].map(result_map)
match_level_df[['opponent', 'result', 'result_points', 'xg_diff']]

correlation = match_level_df['xg_diff'].corr(match_level_df['result_points'])
print(f"Correlation between xG differential and result: {correlation:.3f}")

Correlation between xG differential and result: 0.568

match_level_df['goal_diff'] = match_level_df['goals_for'] - match_level_df['goals_against']
goal_diff_corr = match_level_df['goal_diff'].corr(match_level_df['result_points'])
print(f"Correlation between goal differential and result: {goal_diff_corr:.3f}")

Correlation between goal differential and result: 0.728

without_outlier = match_level_df[match_level_df['opponent'] != 'Bristol City WFC']
corr_no_outlier = without_outlier['xg_diff'].corr(without_outlier['result_points'])
print(f"Correlation without Bristol City match: {corr_no_outlier:.3f}")
print(f"Correlation with Bristol City match: {correlation:.3f}")

Correlation without Bristol City match: 0.561
Correlation with Bristol City match: 0.568

from scipy import stats

wins = match_level_df[match_level_df['result'] == 'W']['xg_diff']
draws = match_level_df[match_level_df['result'] == 'D']['xg_diff']
losses = match_level_df[match_level_df['result'] == 'L']['xg_diff']

print(f"n wins={len(wins)}, n draws={len(draws)}, n losses={len(losses)}")

# Kruskal-Wallis is safer than ANOVA with small/uneven group sizes
h_stat, p_value = stats.kruskal(wins, draws, losses)
print(f"Kruskal-Wallis H-statistic: {h_stat:.3f}, p-value: {p_value:.4f}")

n wins=18, n draws=1, n losses=3
Kruskal-Wallis H-statistic: 6.638, p-value: 0.0362

h_stat_2group, p_value_2group = stats.kruskal(wins, losses)
print(f"Wins vs Losses only — H-statistic: {h_stat_2group:.3f}, p-value: {p_value_2group:.4f}")

Wins vs Losses only — H-statistic: 6.313, p-value: 0.0120

findings = {
    'xg_diff_result_correlation': correlation,
    'goal_diff_result_correlation': goal_diff_corr,
    'kruskal_h_stat_wins_losses': h_stat_2group,
    'kruskal_p_value_wins_losses': p_value_2group,
    'n_wins': len(wins),
    'n_draws': len(draws),
    'n_losses': len(losses),
    'draw_note': 'Only draw was vs Man City, xG diff +1.50 — underperformance relative to underlying numbers',
    'method_note': 'Used wins-vs-losses Kruskal-Wallis rather than 3-group, since n=1 draw is not a statistically meaningful group'
}
print(findings)

import json
with open('session4_findings.json', 'w') as f:
    json.dump(findings, f, indent=2)

{'xg_diff_result_correlation': np.float64(0.5675149273777641), 'goal_diff_result_correlation': np.float64(0.7283377035825139), 'kruskal_h_stat_wins_losses': np.float64(6.313131313131322), 'kruskal_p_value_wins_losses': np.float64(0.011984698922786682), 'n_wins': 18, 'n_draws': 1, 'n_losses': 3, 'draw_note': 'Only draw was vs Man City, xG diff +1.50 — underperformance relative to underlying numbers', 'method_note': 'Used wins-vs-losses Kruskal-Wallis rather than 3-group, since n=1 draw is not a statistically meaningful group'}

# Session 5
import pandas as pd
import numpy as np

match_level_df = pd.read_csv("chelsea_match_level_2023_24.csv")
match_level_df.head()

match_level_df['attack_residual'] = match_level_df['goals_for'] - match_level_df['xg_for']
match_level_df['defense_residual'] = match_level_df['goals_against'] - match_level_df['xg_against']

# A combined "overall luck/performance" measure
match_level_df['net_residual'] = match_level_df['attack_residual'] - match_level_df['defense_residual']

match_level_df[['opponent', 'result', 'goals_for', 'xg_for', 'attack_residual',
                'goals_against', 'xg_against', 'defense_residual', 'net_residual']].round(2)

print("TOP 5 OVERPERFORMANCES (net_residual):")
print(match_level_df.sort_values('net_residual', ascending=False).head(5)[
    ['opponent', 'result', 'goals_for', 'xg_for', 'goals_against', 'xg_against', 'net_residual']
].round(2))

print("\nTOP 5 UNDERPERFORMANCES (net_residual):")
print(match_level_df.sort_values('net_residual').head(5)[
    ['opponent', 'result', 'goals_for', 'xg_for', 'goals_against', 'xg_against', 'net_residual']
].round(2))

TOP 5 OVERPERFORMANCES (net_residual):
               opponent result  goals_for  xg_for  goals_against  xg_against  \
3      Bristol City WFC      W          8    4.56              0        1.41   
15  Manchester United W      W          6    3.30              0        0.37   
20          Everton LFC      W          3    1.16              0        0.98   
16  West Ham United LFC      W          2    1.01              0        1.10   
19        Liverpool WFC      W          5    2.91              1        0.76   

    net_residual  
3           4.85  
15          3.07  
20          2.82  
16          2.09  
19          1.85  

TOP 5 UNDERPERFORMANCES (net_residual):
                      opponent result  goals_for  xg_for  goals_against  \
12  Brighton & Hove Albion WFC      W          4    4.94              2   
14         Manchester City WFC      D          1    2.17              1   
10                 Arsenal WFC      L          1    0.89              4   
4                Liverpool WFC      L          3    1.84              4   
7          Manchester City WFC      L          0    1.47              1   

    xg_against  net_residual  
12        1.39         -1.55  
14        0.67         -1.50  
10        2.71         -1.18  
4         1.68         -1.16  
7         1.45         -1.02

total_attack_residual = match_level_df['attack_residual'].sum()
total_defense_residual = match_level_df['defense_residual'].sum()

print(f"Season total — goals scored above xG: {total_attack_residual:.2f}")
print(f"Season total — goals conceded above xG against: {total_defense_residual:.2f}")

Season total — goals scored above xG: 15.04
Season total — goals conceded above xG against: -3.60

match_level_df['net_residual_zscore'] = (
    (match_level_df['net_residual'] - match_level_df['net_residual'].mean())
    / match_level_df['net_residual'].std()
)

outliers = match_level_df[match_level_df['net_residual_zscore'].abs() > 1.5]
print(f"Matches with |z-score| > 1.5: {len(outliers)}")
print(outliers[['opponent', 'result', 'net_residual', 'net_residual_zscore']].round(2))

Matches with |z-score| > 1.5: 1
           opponent result  net_residual  net_residual_zscore
3  Bristol City WFC      W          4.85                  2.5

outliers_loose = match_level_df[match_level_df['net_residual_zscore'].abs() > 1.0]
print(f"Matches with |z-score| > 1.0: {len(outliers_loose)}")
print(outliers_loose[['opponent', 'result', 'net_residual', 'net_residual_zscore']].round(2))

Matches with |z-score| > 1.0: 8
                      opponent result  net_residual  net_residual_zscore
3             Bristol City WFC      W          4.85                 2.50
4                Liverpool WFC      L         -1.16                -1.25
7          Manchester City WFC      L         -1.02                -1.16
10                 Arsenal WFC      L         -1.18                -1.26
12  Brighton & Hove Albion WFC      W         -1.55                -1.49
14         Manchester City WFC      D         -1.50                -1.46
15         Manchester United W      W          3.07                 1.39
20                 Everton LFC      W          2.82                 1.23

import matplotlib.pyplot as plt

match_level_df_sorted = match_level_df.sort_values('date')
plt.figure(figsize=(10, 5))
plt.bar(range(len(match_level_df_sorted)), match_level_df_sorted['net_residual'],
        color=['green' if x > 0 else 'red' for x in match_level_df_sorted['net_residual']])
plt.axhline(0, color='black', linewidth=0.8)
plt.xlabel('Match (chronological order)')
plt.ylabel('Net Residual (overperformance / underperformance)')
plt.title('Chelsea Women 2023/24: Over/Underperformance vs. xG, By Match')
plt.tight_layout()
plt.show()

match_level_df.to_csv("chelsea_match_level_with_residuals.csv", index=False)

session5_summary = {
    'total_attack_residual': round(total_attack_residual, 2),
    'total_defense_residual': round(total_defense_residual, 2),
    'n_outlier_matches': len(outliers),
    'outlier_opponents': outliers['opponent'].tolist()
}
print(session5_summary)

import json
with open('session5_findings.json', 'w') as f:
    json.dump(session5_summary, f, indent=2)

{'total_attack_residual': np.float64(15.04), 'total_defense_residual': np.float64(-3.6), 'n_outlier_matches': 1, 'outlier_opponents': ['Bristol City WFC']}

#Session 6
import pandas as pd
import numpy as np

match_level_df = pd.read_csv("chelsea_match_level_with_residuals.csv")
match_level_df.head()

import statsmodels.api as sm
import statsmodels.formula.api as smf
from sklearn.metrics import confusion_matrix, classification_report

result_codes = {'L': 0, 'D': 1, 'W': 2}
model_data = match_level_df[['result', 'xg_diff']].copy()
model_data['result_code'] = model_data['result'].map(result_codes)

mn_model = smf.mnlogit('result_code ~ xg_diff', data=model_data).fit()
print(mn_model.summary())

Optimization terminated successfully.
         Current function value: 0.325568
         Iterations 10
                          MNLogit Regression Results                          
==============================================================================
Dep. Variable:            result_code   No. Observations:                   22
Model:                        MNLogit   Df Residuals:                       18
Method:                           MLE   Df Model:                            2
Date:                Thu, 27 Aug 2026   Pseudo R-squ.:                  0.4352
Time:                        20:39:54   Log-Likelihood:                -7.1625
converged:                       True   LL-Null:                       -12.680
Covariance Type:            nonrobust   LLR p-value:                  0.004014
=================================================================================
result_code=1       coef    std err          z      P>|z|      [0.025      0.975]
---------------------------------------------------------------------------------
Intercept        -2.5508      1.967     -1.297      0.195      -6.407       1.305
xg_diff           4.2711      4.057      1.053      0.292      -3.680      12.222
---------------------------------------------------------------------------------
result_code=2       coef    std err          z      P>|z|      [0.025      0.975]
---------------------------------------------------------------------------------
Intercept        -0.2128      1.131     -0.188      0.851      -2.430       2.005
xg_diff           4.5955      3.969      1.158      0.247      -3.183      12.374
=================================================================================

match_level_df['is_win'] = (match_level_df['result'] == 'W').astype(int)

logit_model = smf.logit('is_win ~ xg_diff', data=match_level_df).fit()
print(logit_model.summary())

Optimization terminated successfully.
         Current function value: 0.306144
         Iterations 8
                           Logit Regression Results                           
==============================================================================
Dep. Variable:                 is_win   No. Observations:                   22
Model:                          Logit   Df Residuals:                       20
Method:                           MLE   Df Model:                            1
Date:                Thu, 27 Aug 2026   Pseudo R-squ.:                  0.3543
Time:                        21:00:20   Log-Likelihood:                -6.7352
converged:                       True   LL-Null:                       -10.431
Covariance Type:            nonrobust   LLR p-value:                  0.006552
==============================================================================
                 coef    std err          z      P>|z|      [0.025      0.975]
------------------------------------------------------------------------------
Intercept      0.1110      0.838      0.132      0.895      -1.532       1.754
xg_diff        1.5134      0.811      1.867      0.062      -0.076       3.102
==============================================================================

odds_ratio = np.exp(logit_model.params['xg_diff'])
print(f"Odds ratio for xg_diff: {odds_ratio:.3f}")

Odds ratio for xg_diff: 4.542

match_level_df['predicted_win_prob'] = logit_model.predict(match_level_df)
match_level_df['predicted_win'] = (match_level_df['predicted_win_prob'] > 0.5).astype(int)

print(match_level_df[['opponent', 'result', 'xg_diff', 'predicted_win_prob', 'predicted_win']].round(2))

                      opponent result  xg_diff  predicted_win_prob  \
0                Aston Villa W      W     2.48                0.98   
1                  Everton LFC      W     2.52                0.98   
2          West Ham United LFC      W     1.56                0.92   
3             Bristol City WFC      W     3.15                0.99   
4                Liverpool WFC      L     0.16                0.59   
5      Tottenham Hotspur Women      W     0.87                0.81   
6                  Arsenal WFC      W     1.86                0.95   
7          Manchester City WFC      L     0.02                0.54   
8          Manchester United W      W     0.72                0.77   
9             Bristol City WFC      W     1.78                0.94   
10                 Arsenal WFC      L    -1.82                0.07   
11          Leicester City WFC      W     1.24                0.88   
12  Brighton & Hove Albion WFC      W     3.55                1.00   
13               Aston Villa W      W     4.75                1.00   
14         Manchester City WFC      D     1.50                0.92   
15         Manchester United W      W     2.94                0.99   
16         West Ham United LFC      W    -0.09                0.49   
17          Leicester City WFC      W     2.21                0.97   
18  Brighton & Hove Albion WFC      W     2.16                0.97   
19               Liverpool WFC      W     2.15                0.97   
20                 Everton LFC      W     0.18                0.59   
21     Tottenham Hotspur Women      W     0.48                0.70   

    predicted_win  
0               1  
1               1  
2               1  
3               1  
4               1  
5               1  
6               1  
7               1  
8               1  
9               1  
10              0  
11              1  
12              1  
13              1  
14              1  
15              1  
16              0  
17              1  
18              1  
19              1  
20              1  
21              1

cm = confusion_matrix(match_level_df['is_win'], match_level_df['predicted_win'])
print("Confusion Matrix:")
print(cm)
print()
print(classification_report(match_level_df['is_win'], match_level_df['predicted_win']))

Confusion Matrix:
[[ 1  3]
 [ 1 17]]

              precision    recall  f1-score   support

           0       0.50      0.25      0.33         4
           1       0.85      0.94      0.89        18

    accuracy                           0.82        22
   macro avg       0.68      0.60      0.61        22
weighted avg       0.79      0.82      0.79        22

draw_row = match_level_df[match_level_df['result'] == 'D']
print(draw_row[['opponent', 'xg_diff', 'predicted_win_prob']])

               opponent  xg_diff  predicted_win_prob
14  Manchester City WFC      1.5            0.915372

match_level_df.to_csv("chelsea_match_level_with_predictions.csv", index=False)

session6_summary = {
    'logit_coefficient_xg_diff': round(logit_model.params['xg_diff'], 3),
    'logit_pvalue_xg_diff': round(logit_model.pvalues['xg_diff'], 4),
    'odds_ratio': round(odds_ratio, 3),
    'pseudo_r_squared': round(logit_model.prsquared, 3),
    'model_note': 'Multinomial (3-class W/D/L) model was unstable due to n=1 draw; pivoted to binary Win/Not-Win logistic regression as the primary model'
}
print(session6_summary)

import json
with open('session6_findings.json', 'w') as f:
    json.dump(session6_summary, f, indent=2)

{'logit_coefficient_xg_diff': np.float64(1.513), 'logit_pvalue_xg_diff': np.float64(0.062), 'odds_ratio': np.float64(4.542), 'pseudo_r_squared': np.float64(0.354), 'model_note': 'Multinomial (3-class W/D/L) model was unstable due to n=1 draw; pivoted to binary Win/Not-Win logistic regression as the primary model'}

#Session 7
import pandas as pd
import matplotlib.pyplot as plt

match_level_df = pd.read_csv("chelsea_match_level_with_predictions.csv")
match_level_df['date'] = pd.to_datetime(match_level_df['date'])
match_level_df = match_level_df.sort_values('date').reset_index(drop=True)
match_level_df.head()

fig, ax = plt.subplots(figsize=(9, 7))

colors = match_level_df['result'].map({'W': '#1f77b4', 'D': '#ff7f0e', 'L': '#d62728'})
ax.scatter(match_level_df['xg_for'], match_level_df['goals_for'], c=colors, s=80, edgecolors='black', linewidths=0.5, zorder=3)

max_val = max(match_level_df['xg_for'].max(), match_level_df['goals_for'].max()) + 1
ax.plot([0, max_val], [0, max_val], 'k--', alpha=0.5, label='Expected (Goals = xG)', zorder=1)

# Label the standout matches
standouts = match_level_df[match_level_df['opponent'].isin(['Bristol City WFC', 'Manchester City WFC']) 
                             & (match_level_df['net_residual'].abs() > 1.4)]
for _, row in standouts.iterrows():
    ax.annotate(f"{row['opponent'].replace(' WFC','').replace(' FCW','')}\n({row['result']})",
                (row['xg_for'], row['goals_for']),
                textcoords="offset points", xytext=(8, 5), fontsize=8)

ax.set_xlabel('Expected Goals (xG For)')
ax.set_ylabel('Actual Goals Scored')
ax.set_title("Chelsea Women 2023/24: Actual Goals vs. xG, by Match Result")

from matplotlib.lines import Line2D
legend_elements = [Line2D([0], [0], marker='o', color='w', label='Win', markerfacecolor='#1f77b4', markersize=8),
                   Line2D([0], [0], marker='o', color='w', label='Draw', markerfacecolor='#ff7f0e', markersize=8),
                   Line2D([0], [0], marker='o', color='w', label='Loss', markerfacecolor='#d62728', markersize=8),
                   Line2D([0], [0], color='k', linestyle='--', alpha=0.5, label='Expected (Goals = xG)')]
ax.legend(handles=legend_elements, loc='upper left')

plt.tight_layout()
plt.savefig('chart1_goals_vs_xg.png', dpi=150)
plt.show()

fig, ax = plt.subplots(figsize=(11, 5))

colors2 = ['#2ca02c' if x > 0 else '#d62728' for x in match_level_df['net_residual']]
bars = ax.bar(range(len(match_level_df)), match_level_df['net_residual'], color=colors2, edgecolor='black', linewidth=0.3)

ax.axhline(0, color='black', linewidth=1)
ax.set_xticks(range(len(match_level_df)))
ax.set_xticklabels(match_level_df['opponent'].str.replace(' WFC','').str.replace(' FCW','').str.replace(' Women',''),
                    rotation=75, ha='right', fontsize=8)
ax.set_ylabel('Net Residual (Goal Performance vs. xG)')
ax.set_title('Chelsea Women 2023/24: Match-by-Match Performance vs. Expected Goals')

# Annotate the one true statistical outlier
outlier_idx = match_level_df[match_level_df['opponent'] == 'Bristol City WFC']['net_residual'].idxmax()
ax.annotate('Statistical outlier\n(z = 2.5)', 
            xy=(outlier_idx, match_level_df.loc[outlier_idx, 'net_residual']),
            xytext=(outlier_idx, match_level_df.loc[outlier_idx, 'net_residual'] + 0.5),
            ha='center', fontsize=8, fontweight='bold')

plt.tight_layout()
plt.savefig('chart2_season_timeline.png', dpi=150)
plt.show()

fig, ax = plt.subplots(figsize=(9, 6))

colors3 = match_level_df['is_win'].map({1: '#1f77b4', 0: '#d62728'})
ax.scatter(match_level_df['xg_diff'], match_level_df['predicted_win_prob'], c=colors3, s=70, edgecolors='black', linewidths=0.5)
ax.axhline(0.5, color='gray', linestyle=':', alpha=0.7, label='50% decision threshold')

# Label the Man City draw specifically
draw_row = match_level_df[match_level_df['result'] == 'D'].iloc[0]
ax.annotate('Man City (Draw)\nPredicted 92% win', 
            (draw_row['xg_diff'], draw_row['predicted_win_prob']),
            textcoords="offset points", xytext=(-10, -20), fontsize=9, fontweight='bold',
            arrowprops=dict(arrowstyle='->', lw=1))

ax.set_xlabel('xG Differential')
ax.set_ylabel('Model-Predicted Win Probability')
ax.set_title('Predicted Win Probability vs. xG Differential')

legend_elements = [Line2D([0], [0], marker='o', color='w', label='Actual Win', markerfacecolor='#1f77b4', markersize=8),
                   Line2D([0], [0], marker='o', color='w', label='Actual Non-Win', markerfacecolor='#d62728', markersize=8)]
ax.legend(handles=legend_elements, loc='lower right')

plt.tight_layout()
plt.savefig('chart3_model_fit.png', dpi=150)
plt.show()

	match_date	home_team	away_team	home_score	away_score
5	2024-04-17	Chelsea FCW	Aston Villa W	3	0
10	2024-02-04	Chelsea FCW	Everton LFC	3	0
11	2023-10-14	Chelsea FCW	West Ham United LFC	2	0
18	2024-05-05	Chelsea FCW	Bristol City WFC	8	0
24	2024-05-01	Liverpool WFC	Chelsea FCW	4	3
31	2024-05-15	Tottenham Hotspur Women	Chelsea FCW	0	1
39	2024-03-15	Chelsea FCW	Arsenal WFC	3	1
42	2024-02-16	Chelsea FCW	Manchester City WFC	0	1
45	2024-01-21	Chelsea FCW	Manchester United W	3	1
48	2023-12-17	Bristol City WFC	Chelsea FCW	0	3
52	2023-12-10	Arsenal WFC	Chelsea FCW	4	1
54	2023-11-26	Chelsea FCW	Leicester City WFC	5	2
59	2023-10-22	Chelsea FCW	Brighton & Hove Albion WFC	4	2
60	2023-11-04	Aston Villa W	Chelsea FCW	0	6
70	2023-10-08	Manchester City WFC	Chelsea FCW	1	1
73	2024-05-18	Manchester United W	Chelsea FCW	0	6
84	2024-03-24	West Ham United LFC	Chelsea FCW	0	2
87	2024-03-03	Leicester City WFC	Chelsea FCW	0	4
98	2024-01-27	Brighton & Hove Albion WFC	Chelsea FCW	0	3
115	2023-11-18	Chelsea FCW	Liverpool WFC	5	1
119	2023-11-12	Everton LFC	Chelsea FCW	0	3
127	2023-10-01	Chelsea FCW	Tottenham Hotspur Women	2	1

	match_id	date	opponent	venue	result	goals_for	goals_against	xg_for	xg_against	xg_diff
0	3913165	2024-04-17	Aston Villa W	Home	W	3	0	2.83	0.36	2.48
1	3913134	2024-02-04	Everton LFC	Home	W	3	0	3.08	0.56	2.52
2	3913069	2023-10-14	West Ham United LFC	Home	W	2	0	2.41	0.86	1.56
3	3913180	2024-05-05	Bristol City WFC	Home	W	8	0	4.56	1.41	3.15
4	3913173	2024-05-01	Liverpool WFC	Away	L	3	4	1.84	1.68	0.16
5	3913163	2024-05-15	Tottenham Hotspur Women	Away	W	1	0	1.43	0.56	0.87
6	3913152	2024-03-15	Arsenal WFC	Home	W	3	1	2.49	0.62	1.86
7	3913138	2024-02-16	Manchester City WFC	Home	L	0	1	1.47	1.45	0.02
8	3913121	2024-01-21	Manchester United W	Home	W	3	1	1.70	0.98	0.72
9	3913112	2023-12-17	Bristol City WFC	Away	W	3	0	2.81	1.03	1.78
10	3913105	2023-12-10	Arsenal WFC	Away	L	1	4	0.89	2.71	-1.82
11	3913103	2023-11-26	Leicester City WFC	Home	W	5	2	2.49	1.25	1.24
12	3913077	2023-10-22	Brighton & Hove Albion WFC	Home	W	4	2	4.94	1.39	3.55
13	3913081	2023-11-04	Aston Villa W	Away	W	6	0	5.19	0.43	4.75
14	3912600	2023-10-08	Manchester City WFC	Away	D	1	1	2.17	0.67	1.50
15	3913187	2024-05-18	Manchester United W	Away	W	6	0	3.30	0.37	2.94
16	3913158	2024-03-24	West Ham United LFC	Away	W	2	0	1.01	1.10	-0.09
17	3913145	2024-03-03	Leicester City WFC	Away	W	4	0	2.70	0.49	2.21
18	3913123	2024-01-27	Brighton & Hove Albion WFC	Away	W	3	0	2.30	0.14	2.16
19	3913093	2023-11-18	Liverpool WFC	Home	W	5	1	2.91	0.76	2.15
20	3913089	2023-11-12	Everton LFC	Away	W	3	0	1.16	0.98	0.18
21	3912597	2023-10-01	Tottenham Hotspur Women	Home	W	2	1	2.28	1.80	0.48

	opponent	result	goals_for	xg_for	attack_residual	goals_against	xg_against	defense_residual	net_residual
0	Aston Villa W	W	3	2.83	0.17	0	0.36	-0.36	0.53
1	Everton LFC	W	3	3.08	-0.08	0	0.56	-0.56	0.48
2	West Ham United LFC	W	2	2.41	-0.41	0	0.86	-0.86	0.45
3	Bristol City WFC	W	8	4.56	3.44	0	1.41	-1.41	4.85
4	Liverpool WFC	L	3	1.84	1.16	4	1.68	2.32	-1.16
5	Tottenham Hotspur Women	W	1	1.43	-0.43	0	0.56	-0.56	0.13
6	Arsenal WFC	W	3	2.49	0.51	1	0.62	0.38	0.13
7	Manchester City WFC	L	0	1.47	-1.47	1	1.45	-0.45	-1.02
8	Manchester United W	W	3	1.70	1.30	1	0.98	0.02	1.28
9	Bristol City WFC	W	3	2.81	0.19	0	1.03	-1.03	1.22
10	Arsenal WFC	L	1	0.89	0.11	4	2.71	1.29	-1.18
11	Leicester City WFC	W	5	2.49	2.51	2	1.25	0.75	1.76
12	Brighton & Hove Albion WFC	W	4	4.94	-0.94	2	1.39	0.61	-1.55
13	Aston Villa W	W	6	5.19	0.81	0	0.43	-0.43	1.24
14	Manchester City WFC	D	1	2.17	-1.17	1	0.67	0.33	-1.50
15	Manchester United W	W	6	3.30	2.70	0	0.37	-0.37	3.07
16	West Ham United LFC	W	2	1.01	0.99	0	1.10	-1.10	2.09
17	Leicester City WFC	W	4	2.70	1.30	0	0.49	-0.49	1.79
18	Brighton & Hove Albion WFC	W	3	2.30	0.70	0	0.14	-0.14	0.84
19	Liverpool WFC	W	5	2.91	2.09	1	0.76	0.24	1.85
20	Everton LFC	W	3	1.16	1.84	0	0.98	-0.98	2.82
21	Tottenham Hotspur Women	W	2	2.28	-0.28	1	1.80	-0.80	0.52