河南郑州二手房房价预测

数据集

data_all.info()

data_all.head()

数据标准化

特征的归一化

data_all['unit_price'].astype(float)

train_data = data_all[data_all.columns.delete([0,1,2])]
data_temp = data_all[data_all.columns.delete([0,1])]
# train_data

title_list = [
    'house_region','house_address','house_structure','house_elevator_sytle',
    'house_heating_type','elevator','house_transaction_type','house_useage',
    'house_years','house_floor_position','house_building_structure','house_orientation',
    'house_building_type','house_layout','house_area','house_decoration','house_last_time'
]
for item in title_list:
    pclassDf = pd.DataFrame()
    #使用get_dummies进行one-hot编码，列名前缀是Pclass
    pclassDf = pd.get_dummies( data_all[item] , prefix=item ).astype("int")
    train_data = pd.concat([train_data,pclassDf],axis=1)
    train_data.drop(columns=[item],inplace=True)

    
    data_temp = pd.concat([data_temp,pclassDf],axis=1)
    data_temp.drop(columns=[item],inplace=True)
    

# print(data_all)
# print(train_data)
# print(train_data.head())
print(data_temp.head())

# from sklearn.decomposition import PCA  # 主成分分析法，非监督的机器学习方法

# model = PCA(n_components=1, random_state=25)  # n_components PCA算法中所要保留的主成分个数n，也即保留下来的特征个数n
# # 多变量之间可能存在相关性，从而增加了问题分析的复杂性。
# # 在减少需要分析的指标同时，尽量减少原指标包含信息的损失，以达到对所收集数据进行全面分析的目的。
# house_train = model.fit_transform(house_train)  # fit_transform是fit和transform的组合，既包括了训练又包含了转换。训练PCA模型，同时返回降维后的数据

# model1 = PCA(n_components=1, random_state=25)  # n_components PCA算法中所要保留的主成分个数n，也即保留下来的特征个数n
# house_test_label = model1.fit_transform(house_test_label)
# house_train
# house_test_label

分割数据集和训练集

from sklearn.model_selection import train_test_split

# train_data = data_all.iloc[:, 3:31]
# train_data = data_all[data_all.columns.delete([0,1,2])]
train_label = data_all.iloc[:, 2]
house_train, house_test, house_train_label, house_test_label = train_test_split \
    (train_data, train_label, test_size=0.2, random_state=42)
print("训练数据集：",house_train)
print("训练数据标签：",house_train_label)
print()
print("测试数据集：",house_test)
print("测试数据标签：",house_test_label)
print(data_all.info())

数据归一化

from sklearn.model_selection import train_test_split
from sklearn.preprocessing import MinMaxScaler

# 归一化
# data_temp.info()
scaler_train_data = MinMaxScaler()
scaler_train_data.fit(data_temp)
scaler_train_data_features = scaler_train_data.transform(data_temp)
scaler_train_data_features = pd.DataFrame(scaler_train_data_features)
scaler_train_data_features.columns = data_temp.columns
print(scaler_train_data_features)

train_data = scaler_train_data_features[scaler_train_data_features.columns.delete([0])]
train_label = scaler_train_data_features.iloc[:, 0]
house_train, house_test, house_train_label, house_test_label = train_test_split \
    (train_data, train_label, test_size=0.2, random_state=25)
print("训练数据集：",house_train)
print("训练数据标签：",house_train_label)
print()
print("测试数据集：",house_test)
print("测试数据标签：",house_test_label)

house_test.head()

house_test_label.head()

house_test.info()

预测模型

数据未归一化前的随机森林预测

from sklearn.ensemble import RandomForestRegressor
rfr = RandomForestRegressor() # 随机森林，基于树的方法是不需要进行特征的归一化
rfr.fit(house_train,house_train_label)
rfr.score(house_test,house_test_label)

正确率：

plt.figure(figsize=(12, 6),facecolor='white')
y_pre = rfr.predict(house_test)
plt.scatter(y_pre,house_test_label,marker='o')
plt.scatter(house_test_label,house_test_label)
plt.show()

数据归一化过后的随机森林预测

from sklearn.ensemble import RandomForestRegressor
rfr = RandomForestRegressor() # 随机森林，基于树的方法是不需要进行特征的归一化
rfr.fit(house_train,house_train_label)
rfr.score(house_test,house_test_label)

plt.figure(figsize=(12, 6),facecolor='white')
y_pre = rfr.predict(house_test)
plt.scatter(y_pre,house_test_label,marker='o')
plt.scatter(house_test_label,house_test_label)
plt.show()

SVM径向基核函数预测

r_svr = SVR(kernel="rbf")#径向基核函数
r_svr.fit(house_train,house_train_label)
r_svr.score(house_test,house_test_label)

plt.figure(figsize=(12, 6),facecolor='white')
y_pre = r_svr.predict(house_test)
plt.scatter(y_pre,house_test_label,marker='o')
plt.scatter(house_test_label,house_test_label)
plt.show()

KNN最邻近算法预测

from sklearn.neighbors import KNeighborsRegressor
knn = KNeighborsRegressor(weights="uniform")#K临近回归器
knn.fit(house_train,house_train_label)
knn.score(house_test,house_test_label)

plt.figure(figsize=(12, 6),facecolor='white')
y_pre = knn.predict(house_test)
plt.scatter(y_pre,house_test_label,marker='o')
plt.scatter(house_test_label,house_test_label)
plt.show()

决策树回归预测

from sklearn.tree import DecisionTreeRegressor
#决策树回归
dt = DecisionTreeRegressor()
dt.fit(house_train,house_train_label)
dt.score(house_test,house_test_label)

plt.figure(figsize=(12, 6),facecolor='white')
y_pre = dt.predict(house_test)
plt.scatter(y_pre,house_test_label,marker='o')
plt.scatter(house_test_label,house_test_label)
plt.show()

梯度提升决策分类预测

from sklearn.ensemble import GradientBoostingRegressor#提升树
gbr = GradientBoostingRegressor()
gbr.fit(house_train,house_train_label)
gbr.score(house_test,house_test_label)

plt.figure(figsize=(12, 6),facecolor='white')
y_pre = gbr.predict(house_test)
plt.scatter(y_pre,house_test_label,marker='o')
plt.scatter(house_test_label,house_test_label)
plt.show()

SVM线性核函数预测

from sklearn.svm import SVR
l_svr = SVR(kernel='linear')#线性核函数
l_svr.fit(house_train,house_train_label)
l_svr.score(house_test,house_test_label)
print(l_svr.score(house_test,house_test_label))

plt.figure(figsize=(12, 6),facecolor='white')
y_pre = l_svr.predict(house_test)
plt.scatter(y_pre,house_test_label,marker='o')
plt.scatter(house_test_label,house_test_label)
plt.show()

使用不同的train_size去训练模型

#建立模型用的训练数据集和测试数据集
from sklearn.svm import SVR
from sklearn.ensemble import RandomForestRegressor
size=np.arange(0.6,1,0.1)
scorelist=[[],[],[],[],[],[]]
from sklearn.model_selection import train_test_split
for i in range(0,4):
    
    house_train, house_test, house_train_label, house_test_label = train_test_split \
        (train_data , train_label,train_size=size[i],random_state=5)
    print("size: ",size[i])

    model = SVR(kernel='linear')#线性核函数
    model.fit(house_train,house_train_label)
    print("线性核函数",model.score(house_test,house_test_label))
    scorelist[0].append(model.score(house_test , house_test_label ))
    

    model = RandomForestRegressor() # 随机森林，基于树的方法是不需要进行特征的归一化
    model.fit(house_train,house_train_label)
    model.score(house_test,house_test_label)
    print("随机森林",model.score(house_test,house_test_label))
    scorelist[1].append(model.score(house_test , house_test_label ))
    
    model = SVR(kernel="rbf")#径向基核函数
    model.fit(house_train,house_train_label)
    scorelist[2].append(model.score(house_test , house_test_label ))
    print("径向基核函数",model.score(house_test,house_test_label))
    
    model = KNeighborsRegressor(weights="uniform")#K临近回归器
    model.fit(house_train,house_train_label)
    scorelist[3].append(model.score(house_test , house_test_label ))
    print("K临近回归器",model.score(house_test,house_test_label))
    
    #决策树回归
    model = DecisionTreeRegressor()
    model.fit(house_train,house_train_label)
    scorelist[4].append(model.score(house_test , house_test_label ))
    print("决策树回归",model.score(house_test,house_test_label))
    
    model = GradientBoostingRegressor()
    model.fit(house_train,house_train_label)
    scorelist[5].append(model.score(house_test , house_test_label ))
    print("梯度提升决策分类",model.score(house_test,house_test_label))
    

模型评估

plt.figure(figsize=(12, 6),facecolor='white')

plt.rcParams['font.sans-serif'] = 'SimHei'
plt.rcParams['axes.unicode_minus'] = False
color_list = ('red', 'blue', 'lightgreen', 'cornflowerblue', 'turquoise', 'magenta')
for i in range(0,6):
#     print(size,scorelist[i])
#     print(scorelist[i])
    plt.plot(size,scorelist[i],color=color_list[i])

plt.legend(['SVM线性核函数', '随机森林','SVM径向基核函数','KNN最近邻','决策树回归','梯度提升决策分类'])
plt.xlabel('训练集占比')
plt.ylabel('准确率')
plt.title('不同的模型随着训练集占比变化曲线')
plt.show()

河南郑州二手房房价分析

河南郑州二手房房价分析