kaggle—酒店预订需求预测分析
kaggle—酒店预订需求预测分析（Hotel booking demand）
项目背景：该项目为酒店线上预订业务的研究内容，从酒店运营的角度，分析酒店的房型供给、不同时间段的需求，核心消费群体，影响退订的因素，并建立分类算法模型对酒店订单退订进行预测。
数据来源：kaggle:Hotel booking demand，此项目数据为kaggle上的一个Hotel booking数据集，感兴趣的朋友可以去下载进行练习。
数据介绍：
| 字段名 | 字段含义 |
|---|---|
| hotel | 酒店名 |
| is_canceled | 是否退订 |
| lead_time | 入住时间 |
| arrival_date_year | 入住的年份 |
| arrival_date_month | 入住的月份 |
| arrival_date_week_number | 一年中的第几周 |
| arrival_date_day_of_month | 一年中的第几号 |
| stays_in_weekend_nights | 周末过夜数 |
| stays_in_week_nights | 周中过夜数 |
| adults | 成人数 |
| children | 儿童数 |
| babies | 婴儿数 |
| meal | 订餐情况 |
| country | 国籍 |
| market_segment | 细分市场 |
| distribution_channel | 市场 |
| is_repeated_guest | 是否回头客 |
| previous_cancellations | 客户在预订前取消的预订数量 |
| previous_bookings_not_canceled | 客户在预订之前未取消的预订数量 |
| reserved_room_type | 房型 |
| assigned_room_type | 房间类型编码 |
| booking_changes | 对预订做出的更改数量 |
| deposit_type | 是否交押金 |
| agent | 旅行社id |
| company | 公司 |
| days_in_waiting_list | 确认订单前的审核天数 |
| customer_type | 预订类型 |
| adr | 平均每日放假 |
| required_car_parking_spaces | 客户要求的车位数量 |
| total_of_special_requests | 特殊要求的数量 |
| reservation_status | 订单状态 |
| reservation_status_date | 订单状态的最后设置日期 |
一共包含32个字段，119390条记录。
项目流程
- 数据预处理
- 缺失值处理
- 数据类型转换
- 异常值处理
- 特征工程
- 数值型特征标准化
- 类别型特征 one-hot编码
- 特征选择
- 模型训练
- 模型预测与评估
一、数据预处理
- 导入需要的库
import pandas as pd
from sklearn.model_selection import train_test_split
from sklearn.model_selection import GridSearchCV
from sklearn.tree import DecisionTreeClassifier
from sklearn.ensemble import RandomForestClassifier
import seaborn as sns
- 查看并理解数据
df = pd.read_csv('hotel_bookings.csv',encoding='gbk')
df.head()
df.info()
结果：
RangeIndex: 119390 entries, 0 to 119389
Data columns (total 32 columns):
# Column Non-Null Count Dtype
--- ------ -------------- -----
0 hotel 119390 non-null object
1 is_canceled 119390 non-null int64
2 lead_time 119390 non-null int64
3 arrival_date_year 119390 non-null int64
4 arrival_date_month 119390 non-null object
5 arrival_date_week_number 119390 non-null int64
6 arrival_date_day_of_month 119390 non-null int64
7 stays_in_weekend_nights 119390 non-null int64
8 stays_in_week_nights 119390 non-null int64
9 adults 119390 non-null int64
10 children 119386 non-null float64
11 babies 119390 non-null int64
12 meal 119390 non-null object
13 country 118902 non-null object
14 market_segment 119390 non-null object
15 distribution_channel 119390 non-null object
16 is_repeated_guest 119390 non-null int64
17 previous_cancellations 119390 non-null int64
18 previous_bookings_not_canceled 119390 non-null int64
19 reserved_room_type 119390 non-null object
20 assigned_room_type 119390 non-null object
21 booking_changes 119390 non-null int64
22 deposit_type 119390 non-null object
23 agent 103050 non-null float64
24 company 6797 non-null float64
25 days_in_waiting_list 119390 non-null int64
26 customer_type 119390 non-null object
27 adr 119390 non-null float64
28 required_car_parking_spaces 119390 non-null int64
29 total_of_special_requests 119390 non-null int64
30 reservation_status 119390 non-null object
31 reservation_status_date 119390 non-null object
dtypes: float64(4), int64(16), object(12)
memory usage: 29.1+ MB
发现数据集一共有32个字段，119389行数据，company列有比较明显缺失值，另外arrival_date等表示时间的列需要合并并且转换为日期格式。
- 日期合并及格式转换
由于数据集中的arrival_date_month月份信息为英文表示，先将其转换为中文月份表示，方便后期合并日期
#修改arrival_date_month的英文月份为中文月份
import calendar
month = []
for i in df.arrival_date_month:
mon = list(calendar.month_name).index(i)
month.append(mon)
df.insert(4,"arrival_month",month)
新增一列预订到店的年月日arrival_date,讲原来的年月日拼接
#将年月日拼接
#增加一列预订到店的年月日arrival_date
df[["arrival_date_year","arrival_month","arrival_date_day_of_month"]] = df[["arrival_date_year","arrival_month","arrival_date_day_of_month"]].apply(lambda x:x.astype(str))
date = df.arrival_date_year.str.cat([df.arrival_month,df.arrival_date_day_of_month],".")
df.insert(3,"arrival_date",date)
转换为日期格式
# 转换日期格式
df['arrival_date']=pd.to_datetime(df['arrival_date'])
将原来的年月日信息删除，只采用新建立的arrival_date表示
df.drop(['arrival_date_year','arrival_month','arrival_date_month','arrival_date_week_number'],axis=1,inplace=True)
df.drop(['arrival_date_day_of_month'],axis=1,inplace=True)
- 缺失值处理
#统计缺失值
df.isnull().sum()
#统计缺失率
#df.isnull().sum()/df.shape[0]
结果：
hotel 0
is_canceled 0
lead_time 0
arrival_date_year 0
arrival_date_month 0
arrival_date_week_number 0
arrival_date_day_of_month 0
stays_in_weekend_nights 0
stays_in_week_nights 0
adults 0
children 4
babies 0
meal 0
country **488**
market_segment 0
distribution_channel 0
is_repeated_guest 0
previous_cancellations 0
previous_bookings_not_canceled 0
reserved_room_type 0
assigned_room_type 0
booking_changes 0
deposit_type 0
agent **16340**
company **112593**
days_in_waiting_list 0
customer_type 0
adr 0
required_car_parking_spaces 0
total_of_special_requests 0
reservation_status 0
reservation_status_date 0
dtype: int64
数据的缺失值主要存在于children,country,agent,company4个字段中，缺失最多的是company
一，children缺失4个，且为数值型变量，所以用中位数填充
二，country缺失488个，且为类别型变量，所以使用众数填充
三，agent缺失16340个，缺失率为13.6%，缺失数量较大，但agent表示预订的旅行社，且缺失率小于20%，建议保留，并用0填充，表示没有旅行社ID
四，company缺失112593个，缺失率为94.3%>80%,不具备信息价值有效性，所以直接删除
df.children.fillna(df.children.median(),inplace=True)
df.country.fillna(df.country.mode()[0],inplace=True)
df.agent.fillna(0,inplace=True)
df.drop(['company'],axis=1,inplace=True)
- 异常值处理
通过观察数据集发现，小孩的入住量，旅行社的入住量存在浮点数，数据集中成人，小孩，婴儿字段均为0，即表示该订单入住人数为0，不符合实际。酒店的平均每日消费存在一个大于5000的异常值。我们需要对此进行处理，以免影响到后续的模型建立。
# children、agent字段不可能为浮点数，需修改数据类型
df.children = df.children.astype(int)
df.agent = df.agent.astype(int)
# 根据原数据集介绍，餐饮字段中的Undefined / SC –无餐套餐为一类
df.meal.replace("Undefined", "SC", inplace=True)
#删除异常值的行
zero_guests = list(df["adults"] + df["children"] + df["babies"] == 0)
df.drop(df.index[zero_guests],inplace=True)
#核实adr变量的离群值情况
sns.boxplot(x=df['adr'])
#删除离群值
df = df[df["adr"]<5000]
特征工程
数值型特征标准化处理
由于数值型特征的单位量纲均不一样，模型拟合时容易偏拟合，所以需要做归一化处理，统一量纲，并保留数据规律
- 先将数值型特征提取出来
#数值型特征标准化过程
num_feature = ["lead_time","stays_nights_total","stays_in_weekend_nights","stays_in_week_nights","number_of_people","adults","children","babies","is_repeated_guest","previous_cancellations","previous_bookings_not_canceled","booking_changes","agent","days_in_waiting_list","adr","required_car_parking_spaces","total_of_special_requests"]
- 对数值特征进行标准化处理
from sklearn.preprocessing import StandardScaler
sc_X = StandardScaler()
#df = sc_X.fit_transform(df)
dff=sc_X.fit_transform(df[["lead_time","stays_in_weekend_nights","stays_in_week_nights","adults","children","babies","is_repeated_guest","previous_cancellations","previous_bookings_not_canceled","booking_changes","agent","days_in_waiting_list","adr","required_car_parking_spaces","total_of_special_requests"]])
dff=pd.DataFrame(data=dff, columns=["lead_time","stays_in_weekend_nights","stays_in_week_nights","adults","children","babies","is_repeated_guest","previous_cancellations","previous_bookings_not_canceled","booking_changes","agent","days_in_waiting_list","adr","required_car_parking_spaces","total_of_special_requests"])
类别特征向量化
由于计算机只能识别数值，而不能识别字符串类别信息，所以为了保证信息的完整性，我们需要进行向量化处理，将其转换为模型容易识别的数值型特征
- 提取类别型特征
cat_feature = ["hotel","meal","country","market_segment","distribution_channel","reserved_room_type","assigned_room_type","deposit_type","customer_type"]
- one-hot编码
from sklearn.preprocessing import OneHotEncoder
one_hot=OneHotEncoder()
data_temp=pd.DataFrame(one_hot.fit_transform(df[["hotel","meal","country","market_segment","distribution_channel","reserved_room_type","assigned_room_type","deposit_type","customer_type"]]).toarray(),
columns=one_hot.get_feature_names(["hotel","meal","country","market_segment","distribution_channel","reserved_room_type","assigned_room_type","deposit_type","customer_type"]),dtype='int32')
data_onehot=pd.concat((dff,data_temp),axis=1) #也可以用merge,join
data_onehot.head()
data_onehot['is_canceled'] = df['is_canceled']
降维
在对类别型特征进行one-hot编码后，数据集由原来的32个字段改变为239个字段，维度大大增加，增加了模型训练的时间复杂性以及可能会造成数据分布稀疏的问题。为了更好的训练模型而又尽量不损失太多的数据信息，在此我们使用决策树模型进行特征选择，保留30个特征进行训练。
# 适用于分类模型
from sklearn.tree import DecisionTreeClassifier
from sklearn.feature_selection import RFE
from sklearn.feature_selection import SelectFromModel
def descTree(x,y,n):
# 数据集划分为特征X和标签y,y是分类
X, y = x, y
# 决策树模型
print('使用决策树模型')
tree = DecisionTreeClassifier().fit(X, y)
model1 = SelectFromModel(tree, prefit=True, max_features=n)
d = model1.get_support(indices=True)
print('特征是...')
print(d)
return d
d= descTree(x,y,30)
all_fea = pd.DataFrame(d)
结果：
使用决策树模型
特征是...
[ 0 1 2 3 4 7 8 9 10 11 12 13 14 16 17 19 20 64
72 77 80 156 204 211 214 220 223 232 236 237]
进行特征选择后，我们拿到了所选择的30个特征的索引，我们需要对其进行挑选合并处理做最后的训练数据
fea_num = list(d)
data_stand_fea = x.iloc[:, list(fea_num[:])]
data_stand_fea
模型训练
- 切割数据集
#切割数据集 82开
X_train, X_test, y_train, y_test = train_test_split(data_stand_fea, y, test_size=0.2)
- 采用RandomForest模型进行训练
clf3 = RandomForestClassifier(n_estimators=160,
max_features=0.4,
min_samples_split=2,
n_jobs=-1,
random_state=0)
clf3.fit(X_train,y_train)
- 模型预测与评估
from sklearn.metrics import accuracy_score
y_pred3 = clf3.predict(X_test)
print('The accuracy of prediction is:', accuracy_score(y_test, y_pred3))
随机森林的评分为0.83，作为第一次训练的结果，评分还是不错的，后续我们可以进行模型参数调优或者采用更加复杂的模型进行训练，提高预测精度。在此我们对此进行一个简单的参数调优
- 参数调优
参数调优可以使用GridSearchCV，但在参数数量选择上，不建议太多，否则数据处理量太多，速度会很慢。对应该模型，参数选择"n_estimators"：决策树的量；“max_depth”：决策树的深度（预剪枝）；“max_features”：选择的最大特征量
rf = RandomForestClassifier()
#参数选择
param_dict = {"n_estimators":[100,150,200],"max_depth":[3,5,8,10,15],"max_features":["auto","log2"]}
#网络搜索调优器
rf_model = GridSearchCV(rf,param_grid=param_dict,cv=3)
#模型拟合
CLF = Pipeline(steps=[('preprocessor', preprocessor),
('classifier', rf_model)])
CLF.fit(X_train, y_train)
#不同参数下，最好的评分及其参数
CLF.best_score_
CLF.best_params_