时间序列预测（三）基于Prophet+XGBoost的销售额预测

# 读取数据
raw_data = pd.read_csv('train.csv')
raw_data['datetime'] = raw_data['datetime'].apply(pd.to_datetime)
df_px = raw_data[['datetime', 'count']].copy()
df_px.rename(columns={'datetime': 'ds', 'count': 'y'}, inplace=True)

# 切分数据集
num = 24*14 # 将最后2周划分为测试集
train, test = df_px.iloc[:-num,:], df_px.iloc[-num:,:]

# 模型拟合
m = Prophet(
                growth='linear',
                seasonality_mode='additive',
                interval_width=0.95,
                daily_seasonality=True,
                weekly_seasonality=True,
                yearly_seasonality=False
            )
m.fit(train)

<fbprophet.forecaster.Prophet at 0x7ff4e4fec730>

# 训练集提取preheat特征
predictions_train = m.predict(train.drop('y', axis=1))
# 测试集提取preheat特征
predictions_test = m.predict(test.drop('y', axis=1))
# 合并
predictions = pd.concat([predictions_train, predictions_test], axis=0)

# 将prophet的结果作为特征
df = pd.merge(raw_data, predictions, left_on=['datetime'], right_on=['ds'], how='inner')
df.drop('ds', axis=1, inplace=True)
df.set_index('datetime', inplace=True)

# 构造yhat滞后值
for lag in range(1,6):
    df[f'yhat_lag_{lag}'] = df['yhat'].shift(lag)

# 构造XGBR数据集
X = df.drop('count', axis=1)
y = df['count']

# 切分数据集
X_train, X_test = X.iloc[:-num,:], X.iloc[-num:,:]
y_train, y_test = y.iloc[:-num], y.iloc[-num:]

# 模型拟合
model_xgbr = XGBRegressor(random_state=0)  # 建立XGBR对象
model_xgbr.fit(X_train, y_train)
# 预测结果
pre_y = model_xgbr.predict(X_test)

# 评估指标
model_metrics_functions = [explained_variance_score, mean_absolute_error, mean_squared_error,r2_score]  # 回归评估指标对象集
model_metrics_list = [[m(y_test, pre_y) for m in model_metrics_functions]]  # 回归评估指标列表
regresstion_score = pd.DataFrame(model_metrics_list, index=['model_xgbr'],
                   columns=['explained_variance', 'mae', 'mse', 'r2'])  # 建立回归指标的数据框
regresstion_score  # 模型回归指标

# 模型效果可视化
fig = plt.figure(figsize=(16,6))
plt.title('True and XGBR result comparison', fontsize=20)
plt.plot(y_test, color='red')
plt.plot(pd.Series(pre_y, index=y_test.index), color='green')
plt.xlabel('Hour', fontsize=16)
plt.ylabel('Number of Shared Bikes', fontsize=16)
plt.legend(labels=['True', 'Pre_y'], fontsize=16)
plt.grid()
plt.show()

# 输出特征重要性
features = X.columns  # 获取特征名称
importances = model_xgbr.feature_importances_  # 获取特征重要性

# 通过二维表格形式显示
importances_df = pd.DataFrame()
importances_df['features'] = features
importances_df['importance'] = importances
importances_df = importances_df.sort_values('importance', ascending=False)
importances_df.head(10)