Python实现足球比赛进球数及胜负结果预测任务求助

前置依赖导入 #

import pandas as pd
import numpy as np
import statsmodels.formula.api as smf
from sklearn.tree import DecisionTreeClassifier
from sklearn.preprocessing import LabelEncoder

任务1：泊松计数回归预测两队进球数 #

原有代码错误说明 #

• 变量名不匹配：定义模型为modelArs，调用fit方法时用了不存在的modelManc，直接触发未定义变量报错
• 特征选择错误：将比赛结果变量score2作为自变量输入，属于数据泄露，完全不符合预测逻辑
• 核心特征缺失：没有纳入主客场标识、队伍自身进攻能力、对手防守能力等对进球数影响最大的特征

实现代码 #

# 1. 数据预处理：将单场比赛拆为主、客队两条记录，构造成进球预测长表
home = df[['date', 'team1', 'team2', 'score1']].rename(columns={'team1':'team', 'team2':'opponent', 'score1':'goals'})
home['is_home'] = 1
away = df[['date', 'team1', 'team2', 'score2']].rename(columns={'team2':'team', 'team1':'opponent', 'score2':'goals'})
away['is_home'] = 0
goal_data = pd.concat([home, away], axis=0, ignore_index=True)

# 2. 划分训练集：排除要预测的赛季末场次，此处以2017年5月为分割线可按需调整
train_goal = goal_data[goal_data['date'] < '2017-05-01']

# 3. 拟合泊松回归模型
poisson_model = smf.poisson("goals ~ is_home + C(team) + C(opponent)", data=train_goal)
poisson_result = poisson_model.fit()

# 4. 预测阿森纳（主队）对阵曼联（客队）的进球数
ars_input = pd.DataFrame({'team':['Arsenal'], 'opponent':['Manchester Utd'], 'is_home':[1]})
ars_goals_pred = poisson_result.predict(ars_input).values[0]
manc_input = pd.DataFrame({'team':['Manchester Utd'], 'opponent':['Arsenal'], 'is_home':[0]})
manc_goals_pred = poisson_result.predict(manc_input).values[0]

print(f"阿森纳预测进球数：{round(ars_goals_pred,2)}，曼联预测进球数：{round(manc_goals_pred,2)}")

任务2：决策树预测获胜队伍 #

# 1. 特征工程
dt_data = df.copy()
# 队伍名称编码
le = LabelEncoder()
dt_data['team1_enc'] = le.fit_transform(dt_data['team1'])
dt_data['team2_enc'] = le.transform(dt_data['team2'])
# 计算各队伍赛季场均进球
team_avg_goals = {}
for team in le.classes_:
    home_goals = dt_data[dt_data['team1']==team]['score1'].mean()
    away_goals = dt_data[dt_data['team2']==team]['score2'].mean()
    team_avg_goals[team] = (home_goals + away_goals)/2
dt_data['team1_avg'] = dt_data['team1'].map(team_avg_goals)
dt_data['team2_avg'] = dt_data['team2'].map(team_avg_goals)
# 结果标签编码
dt_data['result_enc'] = dt_data['result'].map({'team1_win':0, 'team2_win':1, 'draw':2})

# 2. 训练决策树模型
X = dt_data[['team1_enc', 'team2_enc', 'team1_avg', 'team2_avg']]
y = dt_data['result_enc']
dt_model = DecisionTreeClassifier(max_depth=3, random_state=42)
dt_model.fit(X, y)

# 3. 预测阿森纳（主队）对阵曼联的结果
pred_input = pd.DataFrame({
    'team1_enc': [le.transform(['Arsenal'])[0]],
    'team2_enc': [le.transform(['Manchester Utd'])[0]],
    'team1_avg': [team_avg_goals['Arsenal']],
    'team2_avg': [team_avg_goals['Manchester Utd']]
})
dt_pred = dt_model.predict(pred_input)[0]
result_map = {0:'阿森纳胜', 1:'曼联胜', 2:'平局'}
print(f"决策树预测结果：{result_map[dt_pred]}")

任务3：mnlogit回归计算比赛结果概率 #

# 1. 拟合多分类logit模型，复用任务2的特征和数据集
mnlogit_model = smf.mnlogit("result_enc ~ team1_enc + team2_enc + team1_avg + team2_avg", data=dt_data)
mnlogit_result = mnlogit_model.fit()

# 2. 预测三类结果的概率
prob_pred = mnlogit_result.predict(pred_input)
print(f"阿森纳获胜概率：{round(prob_pred.iloc[0,0]*100,2)}%")
print(f"曼联获胜概率：{round(prob_pred.iloc[0,1]*100,2)}%")
print(f"平局概率：{round(prob_pred.iloc[0,2]*100,2)}%")

优化建议 #

• 可新增两队历史对阵记录、最近N场比赛滚动进球/失球均值、队伍排名等特征，大幅提升模型准确率
• 泊松回归可替换为双变量泊松模型，解决实际比赛中两队进球事件不独立的问题
• 决策树可调整max_depth、min_samples_split等参数，避免过拟合

内容的提问来源于stack exchange，提问作者larosploc