ch05_Random Forest

Random Forest¶

ⅰ. 모듈 & DATA 특성 확인¶

In [1]:

import numpy as pn
import pandas as pd
import matplotlib.pyplot as plt
import seaborn as sns

In [2]:

member = pd.read_csv("Data/member.csv")
trans = pd.read_csv("Data/transaction.csv")

member.head()

Out[2]:

	id	recency	zip_code	is_referral	channel
0	906145	10	Surburban	0	Phone
1	184478	6	Rural	1	Web
2	394235	7	Surburban	1	Web
3	130152	9	Rural	1	Web
4	940352	2	Urban	0	Web

In [3]:

member.info()

<class 'pandas.core.frame.DataFrame'>
RangeIndex: 64000 entries, 0 to 63999
Data columns (total 6 columns):
 #   Column       Non-Null Count  Dtype 
---  ------       --------------  ----- 
 0   id           64000 non-null  int64 
 1   recency      64000 non-null  int64 
 2   zip_code     64000 non-null  object
 3   is_referral  64000 non-null  int64 
 4   channel      64000 non-null  object
 5   conversion   64000 non-null  int64 
dtypes: int64(4), object(2)
memory usage: 2.9+ MB

In [4]:

member.describe()

Out[4]:

	id	recency	is_referral	conversion
count	64000.000000	64000.000000	64000.000000	64000.000000
mean	550694.137797	5.763734	0.502250	0.146781
std	259105.689773	3.507592	0.499999	0.353890
min	100001.000000	1.000000	0.000000	0.000000
25%	326772.000000	2.000000	0.000000	0.000000
50%	551300.000000	6.000000	1.000000	0.000000
75%	774914.500000	9.000000	1.000000	0.000000
max	999997.000000	12.000000	1.000000	1.000000

In [5]:

trans.head()

Out[5]:

	id	num_item	total_amount
0	906145	5	34000
1	906145	1	27000
2	906145	4	33000
3	184478	4	29000
4	394235	4	33000

In [6]:

trans.info()

<class 'pandas.core.frame.DataFrame'>
RangeIndex: 196836 entries, 0 to 196835
Data columns (total 3 columns):
 #   Column        Non-Null Count   Dtype
---  ------        --------------   -----
 0   id            196836 non-null  int64
 1   num_item      196836 non-null  int64
 2   total_amount  196836 non-null  int64
dtypes: int64(3)
memory usage: 4.5 MB

In [7]:

trans.describe()

Out[7]:

	id	num_item	total_amount
count	196836.000000	196836.000000	196836.000000
mean	550557.552932	3.078365	21837.102969
std	259254.795613	1.478408	8218.005565
min	100001.000000	1.000000	8000.000000
25%	326719.000000	2.000000	15000.000000
50%	550918.000000	3.000000	22000.000000
75%	774916.000000	4.000000	29000.000000
max	999997.000000	6.000000	38000.000000

ⅱ. Data Merging¶

feature engineering

In [8]:

trans['avg_price'] = trans['total_amount'] / trans['num_item']
trans.head()

Out[8]:

	id	num_item	total_amount	avg_price
0	906145	5	34000	6800.0
1	906145	1	27000	27000.0
2	906145	4	33000	8250.0
3	184478	4	29000	7250.0
4	394235	4	33000	8250.0

In [9]:

trans_mean = trans.groupby('id').mean()
trans_cnt = trans['id'].value_counts()

trans_df = pd.concat([trans_mean, trans_cnt], axis=1)
trans_df

Out[9]:

	num_item	total_amount	avg_price	id
100001	3.500000	26000.000000	7500.000000	2
100008	5.000000	26000.000000	5200.000000	1
100032	2.666667	20666.666667	9366.666667	3
100036	3.000000	25800.000000	13273.333333	5
100070	3.250000	21250.000000	8537.500000	4
...	...	...	...	...
999932	5.000000	32000.000000	6400.000000	1
999981	2.000000	22750.000000	12875.000000	4
999990	3.000000	28000.000000	10388.888889	3
999995	2.000000	27000.000000	13500.000000	1
999997	2.000000	13000.000000	6500.000000	1

64000 rows × 4 columns

In [10]:

trans_df.columns = ['num_item', 'total_amount', 'avg_price', 'count']
# trans_df.rename(columns={'id' : 'count'})
trans_df.head()

Out[10]:

	num_item	total_amount	avg_price	count
100001	3.500000	26000.000000	7500.000000	2
100008	5.000000	26000.000000	5200.000000	1
100032	2.666667	20666.666667	9366.666667	3
100036	3.000000	25800.000000	13273.333333	5
100070	3.250000	21250.000000	8537.500000	4

In [11]:

df = member.join(trans_df, on='id')

In [12]:

df

Out[12]:

	id	recency	zip_code	is_referral	channel	conversion	num_item	total_amount	avg_price	count
0	906145	10	Surburban	0	Phone	0	3.333333	31333.333333	14016.666667	3
1	184478	6	Rural	1	Web	0	4.000000	29000.000000	7250.000000	1
2	394235	7	Surburban	1	Web	0	4.000000	20500.000000	5125.000000	2
3	130152	9	Rural	1	Web	0	1.750000	20750.000000	14875.000000	4
4	940352	2	Urban	0	Web	0	3.000000	31000.000000	10333.333333	1
...	...	...	...	...	...	...	...	...	...	...
63995	838295	10	Urban	0	Web	0	3.500000	26000.000000	8012.500000	4
63996	547316	5	Urban	1	Phone	0	1.800000	17800.000000	11300.000000	5
63997	131575	6	Urban	1	Phone	0	4.000000	30500.000000	7833.333333	2
63998	603659	1	Surburban	1	Multichannel	0	3.200000	21600.000000	7583.333333	5
63999	254229	1	Surburban	0	Web	0	3.400000	24400.000000	9280.000000	5

64000 rows × 10 columns

ⅲ. One-hot Encoding¶

In [13]:

df.isna().sum()

Out[13]:

id              0
recency         0
zip_code        0
is_referral     0
channel         0
conversion      0
num_item        0
total_amount    0
avg_price       0
count           0
dtype: int64

In [14]:

df.head()

Out[14]:

	id	recency	zip_code	is_referral	channel	num_item	total_amount	avg_price	count
0	906145	10	Surburban	0	Phone	3.333333	31333.333333	14016.666667	3
1	184478	6	Rural	1	Web	4.000000	29000.000000	7250.000000	1
2	394235	7	Surburban	1	Web	4.000000	20500.000000	5125.000000	2
3	130152	9	Rural	1	Web	1.750000	20750.000000	14875.000000	4
4	940352	2	Urban	0	Web	3.000000	31000.000000	10333.333333	1

In [15]:

print(df['zip_code'].unique(), df['zip_code'].nunique())
print(df['channel'].unique(), df['channel'].nunique())

['Surburban' 'Rural' 'Urban'] 3
['Phone' 'Web' 'Multichannel'] 3

In [16]:

df = pd.get_dummies(df, columns=['zip_code', 'channel'], drop_first=True)

ⅳ. Modeling¶

In [17]:

from sklearn.model_selection import train_test_split

X = df.drop(['id', 'conversion'], axis=1)
y = df['conversion']

X_train, X_test, y_train, y_test = train_test_split(X, y, test_size = 0.3, random_state = 1234)

In [18]:

from sklearn.ensemble import RandomForestClassifier

model = RandomForestClassifier(max_depth = 10, random_state = 1234)
model.fit(X_train, y_train)

Out[18]:

RandomForestClassifier(max_depth=10, random_state=1234)

In [19]:

from sklearn.metrics import accuracy_score, confusion_matrix, classification_report

pred = model.predict(X_test)
print(accuracy_score(y_test, pred))
confusion_matrix(y_test, pred)

0.8713541666666667

Out[19]:

array([[16348,    74],
       [ 2396,   382]], dtype=int64)

In [20]:

print(classification_report(y_test, pred))

              precision    recall  f1-score   support

           0       0.87      1.00      0.93     16422
           1       0.84      0.14      0.24      2778

    accuracy                           0.87     19200
   macro avg       0.85      0.57      0.58     19200
weighted avg       0.87      0.87      0.83     19200

ⅴ. Tree Regressor¶

기본적으로 이진분류모델인 Random Forest를 연속적인 값도 예측이 가능하게 함

In [21]:

from sklearn.ensemble import RandomForestRegressor

rf = RandomForestRegressor(max_depth = 10, random_state = 1234)
rf.fit(X_train, y_train)

Out[21]:

RandomForestRegressor(max_depth=10, random_state=1234)

In [22]:

pred = rf.predict(X_test)
pd.DataFrame([pred, y_test])

# pred의 값은 y_test와는 다르게 0과 1의 형태가 아님
# 주관적인 기준으로 pred의 값을 이진분류형태로 치환

Out[22]:

	0	1	2	3	4	5	6	7	8	9	...	19190	19191	19192	19193	19194	19195	19196	19197	19198	19199
0	0.130254	0.081264	0.121347	0.202229	0.212367	0.136106	0.115631	0.0	0.137646	0.151976	...	0.058867	0.63356	0.086244	0.145602	0.0	0.113222	0.060416	0.132652	0.312275	0.0
1	0.000000	0.000000	0.000000	0.000000	0.000000	0.000000	0.000000	0.0	0.000000	0.000000	...	0.000000	1.00000	0.000000	0.000000	0.0	0.000000	0.000000	0.000000	0.000000	0.0

2 rows × 19200 columns

In [23]:

# pred의 값이 0.5 이상이면 1 반환
res1 = pd.Series(pred).apply(lambda x: 1 if x >= 0.5 else 0)
res1

Out[23]:

0        0
1        0
2        0
3        0
4        0
        ..
19195    0
19196    0
19197    0
19198    0
19199    0
Length: 19200, dtype: int64

In [24]:

print(accuracy_score(y_test, res1))
print(confusion_matrix(y_test, res1))
print(classification_report(y_test, res1))

0.8759375
[[16288   134]
 [ 2248   530]]
              precision    recall  f1-score   support

           0       0.88      0.99      0.93     16422
           1       0.80      0.19      0.31      2778

    accuracy                           0.88     19200
   macro avg       0.84      0.59      0.62     19200
weighted avg       0.87      0.88      0.84     19200

In [25]:

# pred의 값이 0.3 이상이면 1 반환
res2 = pd.Series(pred).apply(lambda x: 1 if x >= 0.3 else 0)
res2

Out[25]:

0        0
1        0
2        0
3        0
4        0
        ..
19195    0
19196    0
19197    0
19198    1
19199    0
Length: 19200, dtype: int64

In [26]:

print(accuracy_score(y_test, res2))
print(confusion_matrix(y_test, res2))
print(classification_report(y_test, res2))

0.8653125
[[15729   693]
 [ 1893   885]]
              precision    recall  f1-score   support

           0       0.89      0.96      0.92     16422
           1       0.56      0.32      0.41      2778

    accuracy                           0.87     19200
   macro avg       0.73      0.64      0.67     19200
weighted avg       0.84      0.87      0.85     19200

ⅵ. 변수 중요도 파악¶

In [27]:

X_train.columns

Out[27]:

Index(['recency', 'is_referral', 'num_item', 'total_amount', 'avg_price',
       'count', 'zip_code_Surburban', 'zip_code_Urban', 'channel_Phone',
       'channel_Web'],
      dtype='object')

In [28]:

rf.feature_importances_

Out[28]:

array([0.04335594, 0.01528215, 0.38190484, 0.14378719, 0.23511524,
       0.1636367 , 0.00359206, 0.0035514 , 0.00596767, 0.00380682])

In [29]:

plt.figure(figsize=(20, 10))
sns.barplot(x=X_train.columns, y=rf.feature_importances_)

Out[29]:

<AxesSubplot:>

저작자표시 비영리 변경금지

'데이터 분석 > Proj. E-Commerce' 카테고리의 다른 글

E-Commerce Part Ⅷ: 자연어분석(NLP) 응용 (0)	2021.01.18
E-Commerce Part Ⅶ: 시계열 분석 응용 (0)	2021.01.18
E-Commerce Part Ⅵ: K Means Clustering 응용 (0)	2021.01.16
E-Commerce Part Ⅳ: Decision Tree 모델 응용 (0)	2021.01.16
E-Commerce Part Ⅲ: KNN 모델 응용 (0)	2021.01.15
E-Commerce Part Ⅱ: 로지스틱회귀분석 응용 (0)	2021.01.15
E-Commerce Part Ⅰ: 선형회귀분석 응용 (0)	2021.01.12

CheeseChaser

E-Commerce Part Ⅴ: Random Forest 모델 응용

Random Forest¶

ⅰ. 모듈 & DATA 특성 확인¶

ⅱ. Data Merging¶

ⅲ. One-hot Encoding¶

ⅳ. Modeling¶

ⅴ. Tree Regressor¶

ⅵ. 변수 중요도 파악¶

'데이터 분석 > Proj. E-Commerce' 카테고리의 다른 글

티스토리툴바

E-Commerce Part Ⅴ: Random Forest 모델 응용

Random Forest¶

ⅰ. 모듈 & DATA 특성 확인¶

ⅱ. Data Merging¶

ⅲ. One-hot Encoding¶

ⅳ. Modeling¶

ⅴ. Tree Regressor¶

ⅵ. 변수 중요도 파악¶

'데이터 분석 > Proj. E-Commerce' 카테고리의 다른 글

'데이터 분석/Proj. E-Commerce' Related Articles

티스토리툴바