机器学习-保险花销预测笔记+代码

读取数据

import numpy as np
import pandas as pd
 
data=pd.read_csv(r'D:\人工智能\python视频\机器学习\5--机器学习-线性回归\5--Lasso回归_Ridge回归_多项式回归\insurance.csv',sep=',')
data.head(n=6)

EDA 数据探索

import matplotlib.pyplot as plt
%matplotlib inline
 
plt.hist(data['charges'])

#上图出现右偏现象，要变成正态分布形式
plt.hist(np.log(data['charges']),bins=20)

特征工程

data=pd.get_dummies(data)
data.head()

x=data.drop('charges',axis=1)
x

y=data['charges']
 
x.fillna(0,inplace=True)
y.fillna(0,inplace=True)
 
from sklearn.model_selection import train_test_split
x_train,x_test,y_train,y_test=train_test_split(x,y,test_size=0.3)
 
from sklearn.preprocessing import StandardScaler
scaler=StandardScaler(with_mean=True,with_std=True).fit(x_train)
 
 
x_train_scaled=scaler.transform(x_train)
x_test_scaled=scaler.transform(x_test)
x_train_scaled

from sklearn.preprocessing import PolynomialFeatures
poly_features=PolynomialFeatures(degree=2,include_bias=False)
x_train_scaled=poly_features.fit_transform(x_train_scaled)
x_test_scaled=poly_features.fit_transform(x_test_scaled)

模型训练

from sklearn.linear_model import LinearRegression
 
 
reg=LinearRegression()
 
reg.fit(x_train_scaled,np.log1p(y_train))
y_predict=reg.predict(x_test_scaled)
 
#%%
from sklearn.linear_model import Ridge
ridge=Ridge(alpha=0.4)
 
ridge.fit(x_train_scaled,np.log1p(y_train))
y_predict_ridge=ridge.predict(x_test_scaled)
#%%
from sklearn.ensemble import GradientBoostingRegressor
booster=GradientBoostingRegressor()
 
booster.fit(x_train_scaled,np.log1p(y_train))
y_predict_booster=ridge.predict(x_test_scaled)

模型评估

from sklearn.metrics import mean_squared_error
 
#log变换之后的
log_rmse_train=np.sqrt(mean_squared_error(y_true=np.log1p(y_train),y_pred=reg.predict(x_train_scaled))) 
log_rmse_test=np.sqrt(mean_squared_error(y_true=np.log1p(y_test),y_pred=y_predict))
#没有做log变换的
rmse_train=np.sqrt(mean_squared_error(y_true=y_train,y_pred=np.exp(reg.predict(x_train_scaled))))
rmse_test=np.sqrt(mean_squared_error(y_true=y_test,y_pred=np.exp(reg.predict(x_test_scaled))))
 
log_rmse_train,log_rmse_test,rmse_train,rmse_test

#log变换之后的
log_rmse_train=np.sqrt(mean_squared_error(y_true=np.log1p(y_train),y_pred=ridge.predict(x_train_scaled))) 
log_rmse_test=np.sqrt(mean_squared_error(y_true=np.log1p(y_test),y_pred=y_predict_ridge))
#没有做log变换的
rmse_train=np.sqrt(mean_squared_error(y_true=y_train,y_pred=np.exp(ridge.predict(x_train_scaled))))
rmse_test=np.sqrt(mean_squared_error(y_true=y_test,y_pred=np.exp(ridge.predict(x_test_scaled))))
 
log_rmse_train,log_rmse_test,rmse_train,rmse_test

#log变换之后的
log_rmse_train=np.sqrt(mean_squared_error(y_true=np.log1p(y_train),y_pred=booster.predict(x_train_scaled))) 
log_rmse_test=np.sqrt(mean_squared_error(y_true=np.log1p(y_test),y_pred=y_predict_booster))
#没有做log变换的
rmse_train=np.sqrt(mean_squared_error(y_true=y_train,y_pred=np.exp(booster.predict(x_train_scaled))))
rmse_test=np.sqrt(mean_squared_error(y_true=y_test,y_pred=np.exp(booster.predict(x_test_scaled))))
 
log_rmse_train,log_rmse_test,rmse_train,rmse_test