Logistic Regression을 통한 고객의 광고 반응률 예측

고객별 광고 반응율 예측¶

Logistic Regression¶

고객의 수입, 연령, 성별 등 데이터를 활용하여
광고에 대한 고객의 반응 여부를 예측한다.

모듈 및 데이터 로드¶

In [1]:

# 모듈 로드

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

In [2]:

# 데이터 로드

data = pd.read_csv('advertising.csv')

데이터 특성 확인¶

In [3]:

data

Out[3]:

	Daily Time Spent on Site	Age	Area Income	Daily Internet Usage	Ad Topic Line	City	Male	Country	Timestamp	Clicked on Ad
0	68.95	NaN	61833.90	256.09	Cloned 5thgeneration orchestration	Wrightburgh	0	Tunisia	3/27/2016 0:53	0
1	80.23	31.0	68441.85	193.77	Monitored national standardization	West Jodi	1	Nauru	4/4/2016 1:39	0
2	69.47	26.0	59785.94	236.50	Organic bottom-line service-desk	Davidton	0	San Marino	3/13/2016 20:35	0
3	74.15	29.0	54806.18	245.89	Triple-buffered reciprocal time-frame	West Terrifurt	1	Italy	1/10/2016 2:31	0
4	68.37	35.0	73889.99	225.58	Robust logistical utilization	South Manuel	0	Iceland	6/3/2016 3:36	0
...	...	...	...	...	...	...	...	...	...	...
995	72.97	30.0	71384.57	208.58	Fundamental modular algorithm	Duffystad	1	Lebanon	2/11/2016 21:49	1
996	51.30	45.0	67782.17	134.42	Grass-roots cohesive monitoring	New Darlene	1	Bosnia and Herzegovina	4/22/2016 2:07	1
997	51.63	51.0	42415.72	120.37	Expanded intangible solution	South Jessica	1	Mongolia	2/1/2016 17:24	1
998	55.55	19.0	41920.79	187.95	Proactive bandwidth-monitored policy	West Steven	0	Guatemala	3/24/2016 2:35	0
999	45.01	26.0	29875.80	178.35	Virtual 5thgeneration emulation	Ronniemouth	0	Brazil	6/3/2016 21:43	1

1000 rows × 10 columns

In [4]:

data.head()

Out[4]:

	Daily Time Spent on Site	Age	Area Income	Daily Internet Usage	Ad Topic Line	City	Male	Country	Timestamp	Clicked on Ad
0	68.95	NaN	61833.90	256.09	Cloned 5thgeneration orchestration	Wrightburgh	0	Tunisia	3/27/2016 0:53	0
1	80.23	31.0	68441.85	193.77	Monitored national standardization	West Jodi	1	Nauru	4/4/2016 1:39	0
2	69.47	26.0	59785.94	236.50	Organic bottom-line service-desk	Davidton	0	San Marino	3/13/2016 20:35	0
3	74.15	29.0	54806.18	245.89	Triple-buffered reciprocal time-frame	West Terrifurt	1	Italy	1/10/2016 2:31	0
4	68.37	35.0	73889.99	225.58	Robust logistical utilization	South Manuel	0	Iceland	6/3/2016 3:36	0

In [5]:

data.tail()

Out[5]:

	Daily Time Spent on Site	Age	Area Income	Daily Internet Usage	Ad Topic Line	City	Male	Country	Timestamp	Clicked on Ad
995	72.97	30.0	71384.57	208.58	Fundamental modular algorithm	Duffystad	1	Lebanon	2/11/2016 21:49	1
996	51.30	45.0	67782.17	134.42	Grass-roots cohesive monitoring	New Darlene	1	Bosnia and Herzegovina	4/22/2016 2:07	1
997	51.63	51.0	42415.72	120.37	Expanded intangible solution	South Jessica	1	Mongolia	2/1/2016 17:24	1
998	55.55	19.0	41920.79	187.95	Proactive bandwidth-monitored policy	West Steven	0	Guatemala	3/24/2016 2:35	0
999	45.01	26.0	29875.80	178.35	Virtual 5thgeneration emulation	Ronniemouth	0	Brazil	6/3/2016 21:43	1

In [6]:

data.sample()

Out[6]:

	Daily Time Spent on Site	Age	Area Income	Daily Internet Usage	Ad Topic Line	City	Male	Country	Timestamp	Clicked on Ad
690	87.16	33.0	60805.93	197.15	Reduced holistic help-desk	Blairville	0	China	2/11/2016 11:50	0

In [7]:

data.shape

Out[7]:

(1000, 10)

In [8]:

data.info()

<class 'pandas.core.frame.DataFrame'>
RangeIndex: 1000 entries, 0 to 999
Data columns (total 10 columns):
 #   Column                    Non-Null Count  Dtype  
---  ------                    --------------  -----  
 0   Daily Time Spent on Site  1000 non-null   float64
 1   Age                       916 non-null    float64
 2   Area Income               1000 non-null   float64
 3   Daily Internet Usage      1000 non-null   float64
 4   Ad Topic Line             1000 non-null   object 
 5   City                      1000 non-null   object 
 6   Male                      1000 non-null   int64  
 7   Country                   1000 non-null   object 
 8   Timestamp                 1000 non-null   object 
 9   Clicked on Ad             1000 non-null   int64  
dtypes: float64(4), int64(2), object(4)
memory usage: 78.2+ KB

In [9]:

data.describe()

Out[9]:

	Daily Time Spent on Site	Age	Area Income	Daily Internet Usage	Male	Clicked on Ad
count	1000.000000	916.000000	1000.000000	1000.000000	1000.000000	1000.00000
mean	65.000200	36.128821	55000.000080	180.000100	0.481000	0.50000
std	15.853615	9.018548	13414.634022	43.902339	0.499889	0.50025
min	32.600000	19.000000	13996.500000	104.780000	0.000000	0.00000
25%	51.360000	29.000000	47031.802500	138.830000	0.000000	0.00000
50%	68.215000	35.000000	57012.300000	183.130000	0.000000	0.50000
75%	78.547500	42.000000	65470.635000	218.792500	1.000000	1.00000
max	91.430000	61.000000	79484.800000	269.960000	1.000000	1.00000

In [10]:

sns.displot(x=data['Area Income'])

Out[10]:

<seaborn.axisgrid.FacetGrid at 0x28679e15ee0>

In [11]:

sns.displot(x=data['Age'])

Out[11]:

<seaborn.axisgrid.FacetGrid at 0x2867a367220>

In [12]:

data['Country'].nunique()

Out[12]:

237

In [13]:

data['City'].nunique()

Out[13]:

969

In [14]:

data['Ad Topic Line'].nunique()

Out[14]:

1000

불필요한 컬럼 제거¶

In [15]:

# Text Data인 Country, City, Ad Topic Line column의 경우 고유값이 너무 많으므로
# 본 프로젝트에서는 제거하여 사용하겠다. 

data.drop(['Country', 'City', 'Ad Topic Line'], axis=1)

Out[15]:

	Daily Time Spent on Site	Age	Area Income	Daily Internet Usage	Male	Timestamp	Clicked on Ad
0	68.95	NaN	61833.90	256.09	0	3/27/2016 0:53	0
1	80.23	31.0	68441.85	193.77	1	4/4/2016 1:39	0
2	69.47	26.0	59785.94	236.50	0	3/13/2016 20:35	0
3	74.15	29.0	54806.18	245.89	1	1/10/2016 2:31	0
4	68.37	35.0	73889.99	225.58	0	6/3/2016 3:36	0
...	...	...	...	...	...	...	...
995	72.97	30.0	71384.57	208.58	1	2/11/2016 21:49	1
996	51.30	45.0	67782.17	134.42	1	4/22/2016 2:07	1
997	51.63	51.0	42415.72	120.37	1	2/1/2016 17:24	1
998	55.55	19.0	41920.79	187.95	0	3/24/2016 2:35	0
999	45.01	26.0	29875.80	178.35	0	6/3/2016 21:43	1

1000 rows × 7 columns

Missing Value 확인 및 처리¶

In [16]:

data.isna().sum() / len(data)

Out[16]:

Daily Time Spent on Site    0.000
Age                         0.084
Area Income                 0.000
Daily Internet Usage        0.000
Ad Topic Line               0.000
City                        0.000
Male                        0.000
Country                     0.000
Timestamp                   0.000
Clicked on Ad               0.000
dtype: float64

Age column의 데이터 중 8.4%가 결측치임을 확인하였다.

In [17]:

data['Age'].mean()

Out[17]:

36.12882096069869

In [18]:

data['Age'].median()

Out[18]:

35.0

In [19]:

# 프로젝트 진행을 위해 Age column의 결측치를 나이의 평균으로 대치(impute)한다.

data = data.fillna(round(data['Age'].mean()))

In [20]:

data

Out[20]:

	Daily Time Spent on Site	Age	Area Income	Daily Internet Usage	Ad Topic Line	City	Male	Country	Timestamp	Clicked on Ad
0	68.95	36.0	61833.90	256.09	Cloned 5thgeneration orchestration	Wrightburgh	0	Tunisia	3/27/2016 0:53	0
1	80.23	31.0	68441.85	193.77	Monitored national standardization	West Jodi	1	Nauru	4/4/2016 1:39	0
2	69.47	26.0	59785.94	236.50	Organic bottom-line service-desk	Davidton	0	San Marino	3/13/2016 20:35	0
3	74.15	29.0	54806.18	245.89	Triple-buffered reciprocal time-frame	West Terrifurt	1	Italy	1/10/2016 2:31	0
4	68.37	35.0	73889.99	225.58	Robust logistical utilization	South Manuel	0	Iceland	6/3/2016 3:36	0
...	...	...	...	...	...	...	...	...	...	...
995	72.97	30.0	71384.57	208.58	Fundamental modular algorithm	Duffystad	1	Lebanon	2/11/2016 21:49	1
996	51.30	45.0	67782.17	134.42	Grass-roots cohesive monitoring	New Darlene	1	Bosnia and Herzegovina	4/22/2016 2:07	1
997	51.63	51.0	42415.72	120.37	Expanded intangible solution	South Jessica	1	Mongolia	2/1/2016 17:24	1
998	55.55	19.0	41920.79	187.95	Proactive bandwidth-monitored policy	West Steven	0	Guatemala	3/24/2016 2:35	0
999	45.01	26.0	29875.80	178.35	Virtual 5thgeneration emulation	Ronniemouth	0	Brazil	6/3/2016 21:43	1

1000 rows × 10 columns

In [21]:

# 모든 결측치를 대치했음을 확인

data.isna().sum()

Out[21]:

Daily Time Spent on Site    0
Age                         0
Area Income                 0
Daily Internet Usage        0
Ad Topic Line               0
City                        0
Male                        0
Country                     0
Timestamp                   0
Clicked on Ad               0
dtype: int64

Train/Test Set 분리¶

In [22]:

from sklearn.model_selection import train_test_split

In [23]:

X = data[['Daily Time Spent on Site','Age','Area Income', 'Daily Internet Usage','Male']]
y = data['Clicked on Ad']

In [24]:

X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.2, random_state=100)

Logistic Regression 모델 만들기¶

In [25]:

from sklearn.linear_model import LogisticRegression

In [26]:

model = LogisticRegression().fit(X_train, y_train)

In [27]:

# Coefficient 확인

model.coef_

Out[27]:

array([[-6.64737762e-02,  2.66015818e-01, -1.15501902e-05,
        -2.44285539e-02,  2.00758165e-03]])

예측 및 평가¶

In [28]:

predictions = model.predict(X_test)

In [29]:

y_test

Out[29]:

249    1
353    0
537    0
424    1
564    1
      ..
684    1
644    0
110    1
28     1
804    1
Name: Clicked on Ad, Length: 200, dtype: int64

In [30]:

from sklearn.metrics import accuracy_score, confusion_matrix

In [31]:

accuracy_score(predictions, y_test)

Out[31]:

0.9

In [32]:

# confusion matrix: array([[TP, FN], [FP, TN]])

confusion_matrix(predictions, y_test)

Out[32]:

array([[92, 12],
       [ 8, 88]], dtype=int64)

참고: 분류 모델의 성능 평가¶

Confusion Matrix = [[TP, FN], [FP, TN]]¶

TP: Actual Positive & Predictive Positive
FN: Actual Positive & Predictive Negative
FP: Actual Negative & Predictive Positive
TN: Actual Negative & Predictive Negative

$Accuracy = \frac{TP + TN}{TP + FN + FP + TN}$

$Precision = \frac{TP}{TP + FP}$

$Recall = \frac{TP}{TP + FN}$

$F1-Score = \frac{2 \times Precision \times Recall}{Precision + Recall}$

Imbalanced Data에서 주로 accuracy 대신 F1 score 사용

In [ ]: