访问量： 1248

工作常用的机器学习预测案例：

  1. 准备用于预测的特征工程（时间、促销、节假日等）
  2. 对分类变量进行 One-Hot 编码
  3. 划分训练集与测试集（按时间切片）
  4. 使用线性回归预测单品销量
  5. 使用 XGBoost 预测不同品类的增长趋势
  6. 使用 LightGBM 构建高维特征的销量预测模型
  7. 使用随机森林预测新品上市后的表现
  8. 构建多输出模型同时预测多个商品的销量
  9. 使用交叉验证评估模型效果（时间序列专用）
  10. 使用SHAP解释模型预测结果（可解释AI）

好的，我们将展示如何进行这些常见的机器学习预测任务。为了演示这些方法，我们需要使用一些常用的数据分析和机器学习库，如 pandas、numpy、scikit-learn、xgboost 和 lightgbm。首先，让我们创建一个示例DataFrame来模拟原始数据，并逐步应用这些机器学习预测任务。

创建示例数据

import pandas as pd
import numpy as np
from sklearn.model_selection import train_test_split, TimeSeriesSplit, cross_val_score
from sklearn.linear_model import LinearRegression
from sklearn.metrics import mean_squared_error
from xgboost import XGBRegressor
from lightgbm import LGBMRegressor
from sklearn.ensemble import RandomForestRegressor
import shap

# 创建示例时间序列数据
dates = pd.date_range(start='2023-01-01', end='2025-06-30', freq='D')
np.random.seed(42)
sales_data = np.cumsum(np.random.normal(loc=100, scale=20, size=len(dates)))

data = {
    'order_date': dates,
    'product_id': np.random.choice(['P{}'.format(i) for i in range(1, 101)], len(dates)),
    'category_code': np.random.choice(['C{}'.format(i) for i in range(1, 11)], len(dates)),
    'amount': sales_data,
    'quantity': np.random.randint(1, 5, size=len(dates)),
    'promotion': np.random.choice([True, False], len(dates))
}

df = pd.DataFrame(data)

# 设置时间为索引
df.set_index('order_date', inplace=True)

print("原始数据:")
print(df.head())

1. 准备用于预测的特征工程（时间、促销、节假日等）

# 添加时间特征
df['year'] = df.index.year
df['month'] = df.index.month
df['day_of_week'] = df.index.dayofweek

# 添加节假日标志
holidays = pd.to_datetime([
    '2023-01-01', '2023-02-22', '2023-04-05', '2023-05-01', '2023-10-01',
    '2024-01-01', '2024-02-10', '2024-04-04', '2024-05-01', '2024-10-01',
    '2025-01-01', '2025-02-19', '2025-04-04', '2025-05-01', '2025-10-01'
])

df['is_holiday'] = df.index.isin(holidays).astype(int)

# 将促销状态转换为数值
df['promotion_numeric'] = df['promotion'].astype(int)

print("\n添加特征后的数据:")
print(df.head())

2. 对分类变量进行 One-Hot 编码

# 对类别代码和产品ID进行One-Hot编码
df_encoded = pd.get_dummies(df, columns=['category_code', 'product_id'])

print("\nOne-Hot编码后的数据:")
print(df_encoded.head())

3. 划分训练集与测试集（按时间切片）

# 按时间顺序划分训练集和测试集
train_size = int(len(df_encoded) * 0.8)
train, test = df_encoded.iloc[:train_size], df_encoded.iloc[train_size:]

X_train = train.drop(columns=['amount'])
y_train = train['amount']
X_test = test.drop(columns=['amount'])
y_test = test['amount']

print("\n训练集和测试集的形状:")
print(X_train.shape, y_train.shape, X_test.shape, y_test.shape)

4. 使用线性回归预测单品销量

# 线性回归模型
lr = LinearRegression()
lr.fit(X_train, y_train)

# 预测
y_pred_lr = lr.predict(X_test)

# 计算均方误差
mse_lr = mean_squared_error(y_test, y_pred_lr)

print("\n线性回归模型 MSE:", mse_lr)

5. 使用 XGBoost 预测不同品类的增长趋势

# XGBoost模型
xgb = XGBRegressor(objective='reg:squarederror', random_state=42)
xgb.fit(X_train, y_train)

# 预测
y_pred_xgb = xgb.predict(X_test)

# 计算均方误差
mse_xgb = mean_squared_error(y_test, y_pred_xgb)

print("\nXGBoost模型 MSE:", mse_xgb)

6. 使用 LightGBM 构建高维特征的销量预测模型

# LightGBM模型
lgb = LGBMRegressor(random_state=42)
lgb.fit(X_train, y_train)

# 预测
y_pred_lgb = lgb.predict(X_test)

# 计算均方误差
mse_lgb = mean_squared_error(y_test, y_pred_lgb)

print("\nLightGBM模型 MSE:", mse_lgb)

7. 使用随机森林预测新品上市后的表现

# 随机森林模型
rf = RandomForestRegressor(random_state=42)
rf.fit(X_train, y_train)

# 预测
y_pred_rf = rf.predict(X_test)

# 计算均方误差
mse_rf = mean_squared_error(y_test, y_pred_rf)

print("\n随机森林模型 MSE:", mse_rf)

8. 构建多输出模型同时预测多个商品的销量

这里我们假设每个商品的销量可以作为一个独立的目标变量。为了简化示例，我们将只选择几个商品进行预测。

# 选择几个商品作为目标变量
target_products = ['P1', 'P2', 'P3']
targets = df[df['product_id'].isin(target_products)].pivot_table(index=df.index, columns='product_id', values='amount', fill_value=0)

# 合并特征和目标变量
df_multi_output = pd.concat([df_encoded.drop(columns=['amount']), targets], axis=1)

# 按时间顺序划分训练集和测试集
train_size = int(len(df_multi_output) * 0.8)
train, test = df_multi_output.iloc[:train_size], df_multi_output.iloc[train_size:]

X_train_multi = train.drop(columns=target_products)
y_train_multi = train[target_products]
X_test_multi = test.drop(columns=target_products)
y_test_multi = test[target_products]

# 多输出随机森林模型
rf_multi = RandomForestRegressor(random_state=42)
rf_multi.fit(X_train_multi, y_train_multi)

# 预测
y_pred_rf_multi = rf_multi.predict(X_test_multi)

# 计算均方误差
mse_rf_multi = mean_squared_error(y_test_multi, y_pred_rf_multi)

print("\n多输出随机森林模型 MSE:", mse_rf_multi)

9. 使用交叉验证评估模型效果（时间序列专用）

# 时间序列交叉验证
tscv = TimeSeriesSplit(n_splits=5)

# 使用XGBoost模型进行交叉验证
cv_scores = cross_val_score(xgb, X_train, y_train, cv=tscv, scoring='neg_mean_squared_error')

# 转换为正数并计算平均值
avg_cv_mse_xgb = -np.mean(cv_scores)

print("\nXGBoost模型交叉验证 MSE 平均值:", avg_cv_mse_xgb)

10. 使用SHAP解释模型预测结果（可解释AI）

# 使用SHAP解释XGBoost模型
explainer = shap.Explainer(xgb)
shap_values = explainer.shap_values(X_test)

# 绘制SHAP摘要图
shap.summary_plot(shap_values, X_test)

综合以上步骤，最终的机器学习预测结果如下：

这段代码展示了从原始数据到经过全面机器学习预测的结果的过程。你可以根据实际需求调整每一步的操作。

import pandas as pd
import numpy as np
from sklearn.model_selection import train_test_split, TimeSeriesSplit, cross_val_score
from sklearn.linear_model import LinearRegression
from sklearn.metrics import mean_squared_error
from xgboost import XGBRegressor
from lightgbm import LGBMRegressor
from sklearn.ensemble import RandomForestRegressor
import shap
import matplotlib.pyplot as plt

# 创建示例时间序列数据
dates = pd.date_range(start='2023-01-01', end='2025-06-30', freq='D')
np.random.seed(42)
sales_data = np.cumsum(np.random.normal(loc=100, scale=20, size=len(dates)))

data = {
    'order_date': dates,
    'product_id': np.random.choice(['P{}'.format(i) for i in range(1, 101)], len(dates)),
    'category_code': np.random.choice(['C{}'.format(i) for i in range(1, 11)], len(dates)),
    'amount': sales_data,
    'quantity': np.random.randint(1, 5, size=len(dates)),
    'promotion': np.random.choice([True, False], len(dates))
}

df = pd.DataFrame(data)

# 设置时间为索引
df.set_index('order_date', inplace=True)

# 添加时间特征
df['year'] = df.index.year
df['month'] = df.index.month
df['day_of_week'] = df.index.dayofweek

# 添加节假日标志
holidays = pd.to_datetime([
    '2023-01-01', '2023-02-22', '2023-04-05', '2023-05-01', '2023-10-01',
    '2024-01-01', '2024-02-10', '2024-04-04', '2024-05-01', '2024-10-01',
    '2025-01-01', '2025-02-19', '2025-04-04', '2025-05-01', '2025-10-01'
])

df['is_holiday'] = df.index.isin(holidays).astype(int)

# 将促销状态转换为数值
df['promotion_numeric'] = df['promotion'].astype(int)

# 对类别代码和产品ID进行One-Hot编码
df_encoded = pd.get_dummies(df, columns=['category_code', 'product_id'])

# 按时间顺序划分训练集和测试集
train_size = int(len(df_encoded) * 0.8)
train, test = df_encoded.iloc[:train_size], df_encoded.iloc[train_size:]

X_train = train.drop(columns=['amount'])
y_train = train['amount']
X_test = test.drop(columns=['amount'])
y_test = test['amount']

# 线性回归模型
lr = LinearRegression()
lr.fit(X_train, y_train)
y_pred_lr = lr.predict(X_test)
mse_lr = mean_squared_error(y_test, y_pred_lr)
print("\n线性回归模型 MSE:", mse_lr)

# XGBoost模型
xgb = XGBRegressor(objective='reg:squarederror', random_state=42)
xgb.fit(X_train, y_train)
y_pred_xgb = xgb.predict(X_test)
mse_xgb = mean_squared_error(y_test, y_pred_xgb)
print("\nXGBoost模型 MSE:", mse_xgb)

# LightGBM模型
lgb = LGBMRegressor(random_state=42)
lgb.fit(X_train, y_train)
y_pred_lgb = lgb.predict(X_test)
mse_lgb = mean_squared_error(y_test, y_pred_lgb)
print("\nLightGBM模型 MSE:", mse_lgb)

# 随机森林模型
rf = RandomForestRegressor(random_state=42)
rf.fit(X_train, y_train)
y_pred_rf = rf.predict(X_test)
mse_rf = mean_squared_error(y_test, y_pred_rf)
print("\n随机森林模型 MSE:", mse_rf)

# 选择几个商品作为目标变量
target_products = ['P1', 'P2', 'P3']
targets = df[df['product_id'].isin(target_products)].pivot_table(index=df.index, columns='product_id', values='amount', fill_value=0)

# 合并特征和目标变量
df_multi_output = pd.concat([df_encoded.drop(columns=['amount']), targets], axis=1)

# 按时间顺序划分训练集和测试集
train_size = int(len(df_multi_output) * 0.8)
train, test = df_multi_output.iloc[:train_size], df_multi_output.iloc[train_size:]

X_train_multi = train.drop(columns=target_products)
y_train_multi = train[target_products]
X_test_multi = test.drop(columns=target_products)
y_test_multi = test[target_products]

# 多输出随机森林模型
rf_multi = RandomForestRegressor(random_state=42)
rf_multi.fit(X_train_multi, y_train_multi)
y_pred_rf_multi = rf_multi.predict(X_test_multi)
mse_rf_multi = mean_squared_error(y_test_multi, y_pred_rf_multi)
print("\n多输出随机森林模型 MSE:", mse_rf_multi)

# 时间序列交叉验证
tscv = TimeSeriesSplit(n_splits=5)
cv_scores = cross_val_score(xgb, X_train, y_train, cv=tscv, scoring='neg_mean_squared_error')
avg_cv_mse_xgb = -np.mean(cv_scores)
print("\nXGBoost模型交叉验证 MSE 平均值:", avg_cv_mse_xgb)

# 使用SHAP解释XGBoost模型
explainer = shap.Explainer(xgb)
shap_values = explainer.shap_values(X_test)

# 绘制SHAP摘要图
shap.summary_plot(shap_values, X_test)