In [1]:

from datetime import datetime
print(f'Päivitetty {datetime.now()}')

Päivitetty 2022-09-11 12:52:08.931765

Luokittelu - kategorisia muuttujia, tasapainotus¶

Data löytyy esimerkiksi lähteestä: https://archive.ics.uci.edu/ml/datasets/bank+marketing

Kohdemuuttuja y sisältää tiedon siitä, onko asiakkaalla määräaikaistalletuksia.

Selvitetään voidaanko asiakastietojen perusteella ennustaa y-muuttujan arvoja.

Mukana on kategorisia muuttujia, jotka täytyy muuntaa dummy-muuttujiksi.

In [2]:

import pandas as pd

# Datan tasapainottamiseen
from imblearn.over_sampling import RandomOverSampler
from imblearn.under_sampling import RandomUnderSampler

from sklearn.model_selection import train_test_split
from sklearn.ensemble import GradientBoostingClassifier
from sklearn.metrics import confusion_matrix, ConfusionMatrixDisplay

# Datan kaikki sarakkeet näkyviin
pd.options.display.max_columns = None

Datan tarkastelua¶

In [3]:

df = pd.read_csv('https://taanila.fi/banking.csv')
df

Out[3]:

	age	job	marital	education	default	housing	loan	contact	month	day_of_week	duration	campaign	pdays	previous	poutcome	emp_var_rate	cons_price_idx	cons_conf_idx	euribor3m	nr_employed	y
0	44	blue-collar	married	basic.4y	unknown	yes	no	cellular	aug	thu	210	1	999	0	nonexistent	1.4	93.444	-36.1	4.963	5228.1	0
1	53	technician	married	unknown	no	no	no	cellular	nov	fri	138	1	999	0	nonexistent	-0.1	93.200	-42.0	4.021	5195.8	0
2	28	management	single	university.degree	no	yes	no	cellular	jun	thu	339	3	6	2	success	-1.7	94.055	-39.8	0.729	4991.6	1
3	39	services	married	high.school	no	no	no	cellular	apr	fri	185	2	999	0	nonexistent	-1.8	93.075	-47.1	1.405	5099.1	0
4	55	retired	married	basic.4y	no	yes	no	cellular	aug	fri	137	1	3	1	success	-2.9	92.201	-31.4	0.869	5076.2	1
...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...
41183	59	retired	married	high.school	unknown	no	yes	telephone	jun	thu	222	1	999	0	nonexistent	1.4	94.465	-41.8	4.866	5228.1	0
41184	31	housemaid	married	basic.4y	unknown	no	no	telephone	may	thu	196	2	999	0	nonexistent	1.1	93.994	-36.4	4.860	5191.0	0
41185	42	admin.	single	university.degree	unknown	yes	yes	telephone	may	wed	62	3	999	0	nonexistent	1.1	93.994	-36.4	4.857	5191.0	0
41186	48	technician	married	professional.course	no	no	yes	telephone	oct	tue	200	2	999	0	nonexistent	-3.4	92.431	-26.9	0.742	5017.5	0
41187	25	student	single	high.school	no	no	no	telephone	may	fri	112	4	999	0	nonexistent	1.1	93.994	-36.4	4.859	5191.0	0

41188 rows × 21 columns

In [4]:

# Kohdemuuttujan jakaumasta nähdään, että dataa kannattaa tasapainottaa
df['y'].value_counts()

Out[4]:

0    36548
1     4640
Name: y, dtype: int64

Muuttujien muunnokset¶

In [5]:

df['education'].unique()

Out[5]:

array(['basic.4y', 'unknown', 'university.degree', 'high.school',
       'basic.9y', 'professional.course', 'basic.6y', 'illiterate'],
      dtype=object)

In [6]:

# education-muuttujan uudelleenkoodaus
df['education'] = df['education'].replace({'basic.4y':'basic', 'basic.6y':'basic', 'basic.9y':'basic'})
df['education'].unique()

Out[6]:

array(['basic', 'unknown', 'university.degree', 'high.school',
       'professional.course', 'illiterate'], dtype=object)

In [7]:

# Kategoriset muuttujat dummy-muuttujiksi
df_dummies = pd.get_dummies(df)
df_dummies.info()

<class 'pandas.core.frame.DataFrame'>
RangeIndex: 41188 entries, 0 to 41187
Data columns (total 62 columns):
 #   Column                         Non-Null Count  Dtype  
---  ------                         --------------  -----  
 0   age                            41188 non-null  int64  
 1   duration                       41188 non-null  int64  
 2   campaign                       41188 non-null  int64  
 3   pdays                          41188 non-null  int64  
 4   previous                       41188 non-null  int64  
 5   emp_var_rate                   41188 non-null  float64
 6   cons_price_idx                 41188 non-null  float64
 7   cons_conf_idx                  41188 non-null  float64
 8   euribor3m                      41188 non-null  float64
 9   nr_employed                    41188 non-null  float64
 10  y                              41188 non-null  int64  
 11  job_admin.                     41188 non-null  uint8  
 12  job_blue-collar                41188 non-null  uint8  
 13  job_entrepreneur               41188 non-null  uint8  
 14  job_housemaid                  41188 non-null  uint8  
 15  job_management                 41188 non-null  uint8  
 16  job_retired                    41188 non-null  uint8  
 17  job_self-employed              41188 non-null  uint8  
 18  job_services                   41188 non-null  uint8  
 19  job_student                    41188 non-null  uint8  
 20  job_technician                 41188 non-null  uint8  
 21  job_unemployed                 41188 non-null  uint8  
 22  job_unknown                    41188 non-null  uint8  
 23  marital_divorced               41188 non-null  uint8  
 24  marital_married                41188 non-null  uint8  
 25  marital_single                 41188 non-null  uint8  
 26  marital_unknown                41188 non-null  uint8  
 27  education_basic                41188 non-null  uint8  
 28  education_high.school          41188 non-null  uint8  
 29  education_illiterate           41188 non-null  uint8  
 30  education_professional.course  41188 non-null  uint8  
 31  education_university.degree    41188 non-null  uint8  
 32  education_unknown              41188 non-null  uint8  
 33  default_no                     41188 non-null  uint8  
 34  default_unknown                41188 non-null  uint8  
 35  default_yes                    41188 non-null  uint8  
 36  housing_no                     41188 non-null  uint8  
 37  housing_unknown                41188 non-null  uint8  
 38  housing_yes                    41188 non-null  uint8  
 39  loan_no                        41188 non-null  uint8  
 40  loan_unknown                   41188 non-null  uint8  
 41  loan_yes                       41188 non-null  uint8  
 42  contact_cellular               41188 non-null  uint8  
 43  contact_telephone              41188 non-null  uint8  
 44  month_apr                      41188 non-null  uint8  
 45  month_aug                      41188 non-null  uint8  
 46  month_dec                      41188 non-null  uint8  
 47  month_jul                      41188 non-null  uint8  
 48  month_jun                      41188 non-null  uint8  
 49  month_mar                      41188 non-null  uint8  
 50  month_may                      41188 non-null  uint8  
 51  month_nov                      41188 non-null  uint8  
 52  month_oct                      41188 non-null  uint8  
 53  month_sep                      41188 non-null  uint8  
 54  day_of_week_fri                41188 non-null  uint8  
 55  day_of_week_mon                41188 non-null  uint8  
 56  day_of_week_thu                41188 non-null  uint8  
 57  day_of_week_tue                41188 non-null  uint8  
 58  day_of_week_wed                41188 non-null  uint8  
 59  poutcome_failure               41188 non-null  uint8  
 60  poutcome_nonexistent           41188 non-null  uint8  
 61  poutcome_success               41188 non-null  uint8  
dtypes: float64(5), int64(6), uint8(51)
memory usage: 5.5 MB

Tasapainotus ja mallien sovittaminen¶

In [8]:

X = df_dummies.drop('y', axis=1)
y = df_dummies['y']

X_train, X_test, y_train, y_test = train_test_split(X, y, random_state=2)

In [9]:

# Sovitetaan gradienttitehostus ilman datan tasapainotusta
gbc1 = GradientBoostingClassifier(max_depth=3, random_state=2)
gbc1.fit(X_train, y_train)

# Tasapainotus RandomOverSampler + gradienttitehostus
# RandomOverSampler kasvattaa satunnaisotoksilla pienempää ryhmää
ros = RandomOverSampler(random_state=2)
X_train2, y_train2 = ros.fit_resample(X_train, y_train)
gbc2 = GradientBoostingClassifier(max_depth=3, random_state=2)
gbc2.fit(X_train2, y_train2)

# Tasapainotus RandomUnderSampler + gradienttitehostus
# RandomUnderSampler pienentää satunnaisotoksilla isompaa ryhmää
rus = RandomUnderSampler(random_state=2)
X_train3, y_train3 = rus.fit_resample(X_train, y_train)
gbc3 = GradientBoostingClassifier(max_depth=3, random_state=2)
gbc3.fit(X_train3, y_train3)

Out[9]:

GradientBoostingClassifier(random_state=2)

Mallien arviointia¶

In [10]:

# Oikeaan osuneiden ennusteiden osuus opetusdatassa
print(f'Ilman tasapainotusta {gbc1.score(X_train, y_train):.3f}')
print(f'RandomOverSampler {gbc2.score(X_train, y_train):.3f}')
print(f'RandomUnderSampler {gbc3.score(X_train, y_train):.3f}')

Ilman tasapainotusta 0.923
RandomOverSampler 0.861
RandomUnderSampler 0.860

In [11]:

# Oikeaan osuneiden ennusteiden osuus testidatassa
print(f'Ilman tasapainostusta {gbc1.score(X_test, y_test):.3f}')
print(f'RandomOverSampler {gbc2.score(X_test, y_test):.3f}')
print(f'RandomUnderSampler {gbc3.score(X_test, y_test):.3f}')

Ilman tasapainostusta 0.915
RandomOverSampler 0.854
RandomUnderSampler 0.854

In [12]:

# Mallien antamat ennusteet testidatalle
y_pred1 = gbc1.predict(X_test)
y_pred2 = gbc2.predict(X_test)
y_pred3 = gbc3.predict(X_test)

In [13]:

# Sekaannusmatriisi ilman tasapainotusta
cm = confusion_matrix(y_test, y_pred1)
ConfusionMatrixDisplay(confusion_matrix=cm).plot()

Out[13]:

<sklearn.metrics._plot.confusion_matrix.ConfusionMatrixDisplay at 0x20860a8ebe0>

In [14]:

# Sekaannusmatriisi (tasapainotus RandomOverSampler)
cm = confusion_matrix(y_test, y_pred2)
ConfusionMatrixDisplay(confusion_matrix=cm).plot()

Out[14]:

<sklearn.metrics._plot.confusion_matrix.ConfusionMatrixDisplay at 0x20863648250>

In [15]:

# Sekaannusmatriisi (tasapainotus RandomUnderSampler)
cm = confusion_matrix(y_test, y_pred3)
ConfusionMatrixDisplay(confusion_matrix=cm).plot()

Out[15]:

<sklearn.metrics._plot.confusion_matrix.ConfusionMatrixDisplay at 0x2086363a760>

Jos tarkoituksena on poimia uudesta datasta niitä, jotka todennäköisimmin tekisivät määräaikaistalletuksia, niin RandomUnderSampler+gradienttitehostus vaikuttaisi parhaalta mallilta. Se tunnistaa testidatasta suurimman osan määräaikaistalletuksia tehneistä.