Predict Credit Card Approval | by Dinesh Chaudhary | Jun, 2024

In right this moment’s digital age, entry to credit score is essential, however manually analyzing bank card purposes could be time-consuming and error-prone. On this mission, I used machine studying to construct a mannequin that predicts bank card approval with excessive accuracy, probably serving to lenders make sooner and extra knowledgeable choices.”

Targets and Method:

• My aim is to construct a mannequin that would predict whether or not an applicant could be accredited for a bank card primarily based on their knowledge. I began by exploring the info, and figuring out patterns and tendencies. Then, I preprocessed the info by dealing with lacking values, and outliers, and performing obligatory transformations. Subsequent, I skilled varied machine studying fashions, together with Logistic Regression, Random Forest, and Gradient Boosting. Lastly, I evaluated the fashions primarily based on their efficiency metrics and chosen one of the best one.”

Achievements and Challenges:

• One key discovering from my knowledge evaluation was that annual revenue, relations, and employment length have been crucial options for predicting approval. I used strategies like SMOTE to handle imbalanced knowledge, the place candidates with excessive danger have been a lot fewer. Curiously, through the financial recession, prioritizing recall (catching good candidates) was extra necessary than precision (avoiding dangerous candidates), so I selected Gradient Boosting as one of the best mannequin primarily based on its recall rating.”

Influence and Expertise:

• My mannequin achieved an accuracy of 88.48%, probably enhancing lenders’ skill to evaluate creditworthiness effectively. This mission allowed me to showcase my abilities in knowledge evaluation, machine studying, and have engineering. I additionally realized priceless classes about moral concerns in AI fashions and the significance of tailoring them to particular enterprise contexts.”

Right here We comply with the Following Steps:

• Exploratory knowledge evaluation(EDA)
• Characteristic engineering
• Characteristic choice
• Knowledge preprocessing
• Mannequin coaching
• Mannequin choice

1.Import all of the required libraries

import numpy as np
import pandas as pd
import matplotlib
import matplotlib.pyplot as plt
import seaborn as sns
import missingno as msno
import warnings
import warnings
from pandas.errors import SettingWithCopyWarning
warnings.simplefilter(motion="ignore", class=SettingWithCopyWarning)
from pathlib import Path
import os
%matplotlib inlinefrom scipy.stats import probplot, chi2_contingency, chi2, stats
from sklearn.model_selection import train_test_split, GridSearchCV, RandomizedSearchCV, cross_val_score, cross_val_predict
from sklearn.base import BaseEstimator, TransformerMixin
from sklearn.pipeline import Pipeline
from sklearn.calibration import CalibratedClassifierCV
from sklearn.compose import ColumnTransformer
from sklearn.preprocessing import OneHotEncoder, MinMaxScaler, OrdinalEncoder
from sklearn.metrics import ConfusionMatrixDisplay, classification_report, roc_curve, roc_auc_score
from imblearn.over_sampling import SMOTE
from sklearn.linear_model import SGDClassifier, LogisticRegression
from sklearn.svm import SVC
from sklearn.tree import DecisionTreeClassifier
from sklearn.ensemble import RandomForestClassifier, GradientBoostingClassifier, BaggingClassifier, AdaBoostClassifier, ExtraTreesClassifier
from sklearn.naive_bayes import GaussianNB
from sklearn.neighbors import KNeighborsClassifier
from sklearn.discriminant_analysis import LinearDiscriminantAnalysis
from sklearn.neural_network import MLPClassifier
from yellowbrick.model_selection import FeatureImportances
import joblib
from sklearn.inspection import permutation_importance
import scikitplot as skplt

2.1.Import dataset

credit_card_data = pd.read_csv("dataset/Credit_card.csv")
credit_card_label_data = pd.read_csv("dataset/Credit_card_label.csv")credit_card_data.head()

credit_card_label_data.head()

2.2.Merge goal variable(‘label’) in authentic dataset

credit_card_full_data = pd.merge(credit_card_data,credit_card_label_data, on = 'Ind_ID')credit_card_full_data.head()

2.3.Rename Characteristic

credit_card_full_data = credit_card_full_data.rename(columns = {'Birthday_count' : 'Age', 'label':'Is_high_risk'})credit_card_full_data.head()

2.4.Cut up the info into coaching and check units

Now we’ll break up the credit_card_data_full_data right into a coaching and testing set. We’ll use 80% of the info for coaching and 20% for testing and retailer them respectively in cc_train_original and cc_test_original variables

# break up the info into prepare and check dataset
def data_split(df, test_size):
train_df, test_df = train_test_split(df, test_size=test_size, random_state=42)
return train_df.reset_index(drop=True), test_df.reset_index(drop=True)# moist set check measurement 0.2, which suggests our coaching knowledge set is 0.8
cc_train_original, cc_test_original = data_split(credit_card_full_data, 0.2)

Form of coaching knowledge

cc_train_original.form(1238, 19)

• Right here we’ve 19 options(columns) and 1238 observations(rows) for the coaching dataset.

Form of coaching knowledge

cc_test_original.form(310, 19)

• Right here we’ve 19 options(columns) and 310 observations(rows) for the testing dataset.

Save prepare and check knowledge

cc_train_original.to_csv('dataset/prepare.csv',index=False)
cc_test_original.to_csv('dataset/check.csv',index=False)

3.1.Exploring the info

credit_card_full_data.head()

Get Options

credit_card_full_data.columnsIndex(['Ind_ID', 'GENDER', 'Car_Owner', 'Propert_Owner', 'CHILDREN',
'Annual_income', 'Type_Income', 'EDUCATION', 'Marital_status',
'Housing_type', 'Age', 'Employed_days', 'Mobile_phone', 'Work_Phone',
'Phone', 'EMAIL_ID', 'Type_Occupation', 'Family_Members',
'Is_high_risk'],
dtype='object')

Get Information

credit_card_full_data.data()<class 'pandas.core.body.DataFrame'>
RangeIndex: 1548 entries, 0 to 1547
Knowledge columns (complete 19 columns):
#   Column           Non-Null Depend  Dtype  
---  ------           --------------  -----  
0   Ind_ID           1548 non-null   int64  
1   GENDER           1541 non-null   object 
2   Car_Owner        1548 non-null   object 
3   Propert_Owner    1548 non-null   object 
4   CHILDREN         1548 non-null   int64  
5   Annual_income    1525 non-null   float64
6   Type_Income      1548 non-null   object 
7   EDUCATION        1548 non-null   object 
8   Marital_status   1548 non-null   object 
9   Housing_type     1548 non-null   object 
10  Age              1526 non-null   float64
11  Employed_days    1548 non-null   int64  
12  Mobile_phone     1548 non-null   int64  
13  Work_Phone       1548 non-null   int64  
14  Cellphone            1548 non-null   int64  
15  EMAIL_ID         1548 non-null   int64  
16  Type_Occupation  1060 non-null   object 
17  Family_Members   1548 non-null   int64  
18  Is_high_risk     1548 non-null   int64  
dtypes: float64(2), int64(9), object(8)
reminiscence utilization: 229.9+ KB

Describes

• The describe() operate offers us statistics in regards to the numerical options within the dataset. These statistics embrace every numerical characteristic’s depend, imply, customary deviation, interquartile vary(25%, 50%, 75%), and minimal and most values

credit_card_full_data.describe()

Lacking Worth

• Discover lacking worth
• Additionally we’re utilizing Missingno to visualise the lacking values per characteristic utilizing its matrix operate

credit_card_full_data.isnull().sum()Ind_ID               0
GENDER               7
Car_Owner            0
Propert_Owner        0
CHILDREN             0
Annual_income       23
Type_Income          0
EDUCATION            0
Marital_status       0
Housing_type         0
Age                 22
Employed_days        0
Mobile_phone         0
Work_Phone           0
Cellphone                0
EMAIL_ID             0
Type_Occupation    488
Family_Members       0
Is_high_risk         0
dtype: int64msno.matrix(credit_card_full_data)
plt.present()

• Right here we are able to see that the GENDER, Annual_income, Age and Tyoe_Occupation options with lacking values. Slim white traces characterize lacking values

Bar Plot

• we are able to additionally use bar() operate to have a barplot with the depend of non-null values for clear illustration of the lacking values depend

msno.bar(credit_card_full_data)
plt.present()

3.2.create a features to research every characteristic(Univariate evaluation)

3.2.1.Worth Depend Freq operate

• we’ll create first operate value_cnt_freq is used to calculate the depend of every class in a characteristic with its frequency (normalized on a scale of 100)

def value_cnt_freq(df, characteristic):
# we get the worth counts of every characteristic
ftr_value_cnt = df[feature].value_counts()
# we normalize the worth counts on a scale of 100
ftr_value_cnt_norm = df[feature].value_counts(normalize=True) * 100
# we concatenate the worth counts with normalized worth depend column sensible
ftr_value_cnt_concat = pd.concat([ftr_value_cnt, ftr_value_cnt_norm], axis=1)
# give it a column identify
ftr_value_cnt_concat.columns = ['Count', 'Frequency (%)']
# return the dataframe
return ftr_value_cnt_concat

3.2.2.Characteristic Information operate

• Create feature_info operate that may return the outline, the datatype, statistics, the worth counts and frequencies.

def feature_info(df, characteristic):
# if the characteristic is Age
if characteristic == 'Age':
# change the characteristic in constructive variety of days and divide by 365.25 to transform in 12 months
print('Description:n{}'.format((np.abs(df[feature])/365.25).describe()))
print('*'*40)
print('Knowledge sort:{}'.format(df[feature].dtype))
# if the characteristic is Employed_days 
if characteristic == 'Employed_days':
employment_days_no_ret = df['Employed_days'][df['Employed_days'] < 0]
# change destructive to constructive values
employment_days_no_ret_yrs = np.abs(employment_days_no_ret) /365.25
print('Description:n{}'.format((employment_days_no_ret_yrs).describe()))
print('*'*40)
# print the datatype
print('Knowledge sort:{}'.format(employment_days_no_ret.dtype))
else:
# get the outline
print('Description:n{}'.format(df[feature].describe()))
# print separators
print('*'*40)
# print the datatype
print('Knowledge sort:n{}'.format(df[feature].dtype))
# print separators
print('*'*40)
# calling the value_cnt_freq operate
value_cnt = value_cnt_freq(df,characteristic)
# print the outcome
print('Worth depend:n{}'.format(value_cnt))

3.2.3.Bar Plot Perform

• We’re making a operate right here to generate a bar plot.

def bar_plot(df,characteristic):
if characteristic == 'Marital_status' or characteristic == 'Housing_type' or characteristic == 'Type_Occupation' or characteristic == 'Type_Income' or characteristic == 'EDUCATION':
fig, ax = plt.subplots(figsize=(6,10))
sns.barplot(x=value_cnt_freq(df,characteristic).index,y=value_cnt_freq(df,characteristic).values[:,0])
# set the plot's tick labels to the index from the value_cnt_norm_cal operate, rotate these ticks by 45 levels
ax.set_xticks(vary(len(value_cnt_freq(df, characteristic).index)))
ax.set_xticklabels(labels = value_cnt_freq(df,characteristic).index,rotation=45,ha='proper')
# Give the X-axis the identical label because the characteristic identify
plt.xlabel('{}'.format(characteristic))
# Give the Y-axis the label "Depend"
plt.ylabel('Depend')
# Give the plot a title
plt.title('{} depend'.format(characteristic))
# Return the title
return plt.present()
else:
fig, ax = plt.subplots(figsize=(6,10))
sns.barplot(x=value_cnt_freq(df,characteristic).index,y=value_cnt_freq(df,characteristic).values[:,0])
plt.xlabel('{}'.format(characteristic))
plt.ylabel('Depend')
plt.title('{} depend'.format(characteristic))
return plt.present()

3.2.4. Pie Chart Perform

• We’re making a operate to generate a pie chart plot

def pie_chart_plot(df, characteristic):
if characteristic == 'Housing_type' or characteristic == 'EDUCATION':
# name value_cnt_freq operate
ratio_size = value_cnt_freq(df, characteristic)
# get what number of lessons we've
ratio_size_len = len(ratio_size.index)
ratio_list = []
# loop continues till the max vary
for i in vary(ratio_size_len):
# append the ratio of every characteristic to the listing
ratio_list.append(ratio_size.iloc[i]['Frequency (%)'])
# create subplot
fig, ax = plt.subplots(figsize=(6, 6))
plt.pie(ratio_list, startangle=90, autopct='%1.2f%%', wedgeprops={'edgecolor': 'black'})
# add title
plt.title('Pie chart of {}'.format(characteristic))
# add legend
plt.legend(loc='finest', labels=ratio_size.index)
# middle the plot within the subplot
plt.axis('equal')
return plt.present()
# For different options
else:
ratio_size = value_cnt_freq(df, characteristic)
ratio_size_len = len(ratio_size.index)
ratio_list = []
for i in vary(ratio_size_len):
ratio_list.append(ratio_size.iloc[i]['Frequency (%)'])
# create subplot    
fig, ax = plt.subplots(figsize=(8, 8))
plt.pie(ratio_list, labels=ratio_size.index, autopct='%1.2f%%', startangle=90, wedgeprops={'edgecolor': 'black'})
# add title
plt.title('Pie chart of {}'.format(characteristic))
# add legend
plt.legend(loc='finest')
# axis equal to make sure that pie is drawn as a circle.
plt.axis('equal')
# return the plot
return plt.present()

3.2.5.Field-Plot Perform

• We’re making a operate for producing a field plot.

def box_plot(df,characteristic):
if characteristic == 'Age':
fig, ax = plt.subplots(figsize = (2,8))
# convert the characteristic in constructive numbers days
sns.boxplot(y = np.abs(df[feature])/365.25)
plt.title('{} distribution(Boxplot)'.format(characteristic))
return plt.present()
if characteristic == 'CHILDREN':
fig, ax = plt.subplots(figsize=(2,8))
sns.boxplot(y=df[feature])
plt.title('{} distribution(Boxplot)'.format(characteristic))
# utilizing the numpy prepare to populate the Y ticks ranging from 0 until the max depend of kids with an interval of 1 
# as follows np.arange(begin, cease, step)
plt.yticks(np.arange(0,df[feature].max(),1))
return plt.present()
if characteristic == 'Employed_days':
fig, ax = plt.subplots(figsize=(2,8))
employment_days_no_ret = df['Employed_days'][df['Employed_days'] < 0]
# change destructive to constructive values
employment_days_no_ret_yrs = np.abs(employment_days_no_ret) /365.25
# create boxplot
sns.boxplot(y = employment_days_no_ret_yrs)
plt.title('{} distribution(Boxplot)'.format(characteristic))
# set y-axis ticks from 0 to the max worth with an interval of two
plt.yticks(np.arange(0,employment_days_no_ret_yrs.max(),2))
return plt.present()
if characteristic == 'Annual_income':
fig, ax = plt.subplots(figsize=(2,8))
sns.boxplot(y=df[feature])
plt.title('{} distribution(Boxplot)'.format(characteristic))
# format y-axis ticks as integers with commas
ax.get_yaxis().set_major_formatter(
matplotlib.ticker.FuncFormatter(lambda x, p: format(int(x), ',')))
return plt.present()
else:
fig, ax = plt.subplots(figsize=(2,8))
sns.boxplot(y=df[feature])
plt.title('{} distribution(Boxplot)'.format(characteristic))
return plt.present()

3.2.6.Histogram Perform

• We’re making a operate to generate a histogram.

def hist_plot(df, characteristic, the_bins = 50):
if characteristic == 'Age':
fig, ax = plt.subplots(figsize = (18,10))
# convert the characteristic in constructive numbers of days
sns.histplot(np.abs(df[feature])/365.25,bins=the_bins,kde=True)
plt.title('{} distribution'.format(characteristic))
return plt.present()
elif characteristic == 'Annual_income':
fig, ax = plt.subplots(figsize=(18,10))
sns.histplot(df[feature], bins = the_bins, kde = True)
ax.get_xaxis().set_major_formatter(
matplotlib.ticker.FuncFormatter(lambda x, p: format(int(x), ',')))
plt.title('{} distribution'.format(characteristic))
return plt.present()
elif characteristic == 'Employed_days':
fig, ax = plt.subplots(figsize=(18,10))
employment_days_no_ret = df['Employed_days'][df['Employed_days'] < 0]
# change destructive to constructive values
employment_days_no_ret_yrs = np.abs(employment_days_no_ret) /365.25
sns.histplot(employment_days_no_ret_yrs,bins=the_bins,kde=True)
plt.title('{} distribution'.format(characteristic))
return plt.present()
else:
fig, ax = plt.subplots(figsize=(18,10))
sns.histplot(df[feature],bins=the_bins,kde=True)
plt.title('{} distribution'.format(characteristic))
return plt.present()

3.2.7.Low vs Excessive Danger Field Plot Perform

• This operate will plot two field plots: one for low-risk (good shopper) candidates and the opposite for high-risk (dangerous shopper) candidates.

def low_vs_high_risk_box_plot(df, characteristic):
if characteristic == 'Age':
print(np.abs(df.groupby('Is_high_risk')[feature].imply()/365.25))
fig, ax = plt.subplots(figsize=(5,8))
sns.boxplot(y=np.abs(df[feature])/365.25,x=df['Is_high_risk'])
# add ticks to the X axis
plt.xticks(ticks=[0,1],labels=['no','yes'])
plt.title('Excessive danger grouped by age')
return plt.present()
if characteristic == 'Annual_income':
print(np.abs(df.groupby('Is_high_risk')[feature].imply()))
fig, ax = plt.subplots(figsize=(5,8))
sns.boxplot(y=np.abs(df[feature]),x=df['Is_high_risk'])
plt.xticks(ticks=[0,1],labels=['no','yes'])
ax.get_yaxis().set_major_formatter(
matplotlib.ticker.FuncFormatter(lambda x, p: format(int(x), ',')))
plt.title('Excessive danger grouped by {}'.format(characteristic))
return plt.present() 
if characteristic == 'Employed_days':
# test an applicant is excessive danger or not
employment_no_ret = cc_train_original['Employed_days'][cc_train_original['Employed_days'] <0]
employment_no_ret_idx = employment_no_ret.index
employment_days_no_ret_yrs = np.abs(employment_no_ret)/365.25
# extract those that are employed from the unique dataframe and return solely the employment size and Is excessive danger columns
employment_no_ret_df = cc_train_original.iloc[employment_no_ret_idx][['Employed_days','Is_high_risk']]
# return the imply employment size group by how dangerous is the applicant
employment_no_ret_is_high_risk = employment_no_ret_df.groupby('Is_high_risk')['Employed_days'].imply()
print(np.abs(employment_no_ret_is_high_risk)/365.25)
fig, ax = plt.subplots(figsize=(5,8))
sns.boxplot(y= employment_days_no_ret_yrs,x=df['Is_high_risk'])
plt.xticks(ticks=[0,1],labels=['no','yes'])
plt.title('Excessive vs low danger  grouped by {}'.format(characteristic))
return plt.present()
else:
print(np.abs(df.groupby('Is_high_risk')[feature].imply()))
fig, ax = plt.subplots(figsize=(5,8))
sns.boxplot(y=np.abs(df[feature]),x=df['Is_high_risk'])
plt.xticks(ticks=[0,1],labels=['no','yes'])
plt.title('Excessive danger grouped by {}'.format(characteristic))
return plt.present()

3.2.8.Low vs Excessive Danger Bar Plot

• We’re making a operate for a bar plot evaluating Low and Excessive classes.

def low_high_risk_bar_plot(df, characteristic):
# Group by the desired characteristic and sum the high-risk candidates
is_high_risk_grp = df.groupby(characteristic)['Is_high_risk'].sum()
# Kind in descending order
is_high_risk_grp_srt = is_high_risk_grp.sort_values(ascending=False)
# Create a bar plot
fig, ax = plt.subplots(figsize=(6, 10))
sns.barplot(x=is_high_risk_grp_srt.index, y=is_high_risk_grp_srt.values)
# Set the ticks and labels after creating the plot
ax.set_xticks(vary(len(is_high_risk_grp_srt)))
ax.set_xticklabels(labels=is_high_risk_grp_srt.index, rotation=45, ha='proper')
plt.ylabel('Depend')
plt.title(f'Excessive danger candidates depend grouped by {characteristic}')
plt.present()
# Instance utilization:
# low_high_risk_bar_plot(your_dataframe, 'Marital_status')

Age

feature_info(cc_train_original, 'Age')Description:
depend    1220.000000
imply       43.863197
std        11.521349
min        22.422998
25%        33.952772
50%        42.945927
75%        53.479808
max        68.298426
Title: Age, dtype: float64
****************************************
Knowledge sort:float64
Description:
depend     1220.000000
imply    -16021.032787
std       4208.172710
min     -24946.000000
25%     -19533.500000
50%     -15686.000000
75%     -12401.250000
max      -8190.000000
Title: Age, dtype: float64
****************************************
Knowledge sort:
float64
****************************************
Worth depend:
Depend  Frequency (%)
Age                           
-21363.0      5       0.409836
-13557.0      5       0.409836
-18173.0      4       0.327869
-14523.0      4       0.327869
-13682.0      3       0.245902
...         ...            ...
-14661.0      1       0.081967
-8304.0       1       0.081967
-18661.0      1       0.081967
-15429.0      1       0.081967
-18348.0      1       0.081967
[1046 rows x 2 columns]

• we are able to see that the youngest candidates is 22 12 months previous whereas the oldest is 68.and common age is 43.86 and median age is 42.94

box_plot(cc_train_original,'Age')

hist_plot(cc_train_original,'Age')

• we plot histogrma with the kernal density estimator. we are able to see ‘Age’ will not be usually distributed.it’s sligtly positively skewed.

low_vs_high_risk_box_plot(cc_train_original, 'Age')Is_high_risk
0    43.733156
1    44.834892
Title: Age, dtype: float64

• we are able to see that there isn’t any significance distinction between the age of those that are excessive danger and those that will not be. The imply age for each group is round 43 12 months previous and there’s no correlation between the age and danger elements of the candidates.

Gender

feature_info(cc_train_original, 'GENDER')Description:
depend     1232
distinctive       2
prime          F
freq       778
Title: GENDER, dtype: object
****************************************
Knowledge sort:
object
****************************************
Worth depend:
Depend  Frequency (%)
GENDER                      
F         778      63.149351
M         454      36.850649

• we are able to see which have two distinctive lessons F(Feminine) and M(Male) with 778 and 454 repectively. And have 63.14% feminine and 36.85% Male

bar_plot(cc_train_original,'GENDER')

pie_chart_plot(cc_train_original,'GENDER')

Marital_status

feature_info(cc_train_original,'Marital_status')Description:
depend        1238
distinctive          5
prime       Married
freq          848
Title: Marital_status, dtype: object
****************************************
Knowledge sort:
object
****************************************
Worth depend:
Depend  Frequency (%)
Marital_status                            
Married                 848      68.497577
Single / not married    183      14.781906
Civil marriage           76       6.138934
Separated                71       5.735057
Widow                    60       4.846527pie_chart_plot(cc_train_original,'Marital_status')

bar_plot(cc_train_original,'Marital_status')

low_high_risk_bar_plot(cc_train_original,'Marital_status')

• Right here we are able to see there are 5 distinctive lessons.Most candidates belong to married with 68.50 % .
• fascinating remark we are able to see that although we’ve the next variety of candidates who’re separated than widows.however widow candidates are excessive danger than those that are separated.

Family_Members

feature_info(cc_train_original,'Family_Members')Description:
depend    1238.000000
imply        2.172052
std         0.968524
min         1.000000
25%         2.000000
50%         2.000000
75%         3.000000
max        15.000000
Title: Family_Members, dtype: float64
****************************************
Knowledge sort:
int64
****************************************
Worth depend:
Depend  Frequency (%)
Family_Members                      
2                 642      51.857835
1                 264      21.324717
3                 214      17.285945
4                 102       8.239095
5                  14       1.130856
15                  1       0.080775
6                   1       0.080775box_plot(cc_train_original,'Family_Members')

bar_plot(cc_train_original,'Family_Members')

• we are able to see that is numerical characteristic with the median of two relations with 51.85%(642) of all , adopted by single member of the family with 21.32%(264)
• In field plot we are able to see there are three outlier 5 and two are extream with 6 and 15 members.

CHILDREN

feature_info(cc_train_original,'CHILDREN')Description:
depend    1238.000000
imply        0.421648
std         0.804142
min         0.000000
25%         0.000000
50%         0.000000
75%         1.000000
max        14.000000
Title: CHILDREN, dtype: float64
****************************************
Knowledge sort:
int64
****************************************
Worth depend:
Depend  Frequency (%)
CHILDREN                      
0           869      70.193861
1           245      19.789984
2           107       8.642973
3            15       1.211632
14            1       0.080775
4             1       0.080775box_plot(cc_train_original,'CHILDREN')

bar_plot(cc_train_original,'CHILDREN')

• Right here we are able to see most candidates don’t have a toddler
• Additionally right here have three outlier 3 and extream outlier with 4 and 15

Housing_type

feature_info(cc_train_original,'Housing_type')Description:
depend                  1238
distinctive                    6
prime       Home / condo
freq                   1107
Title: Housing_type, dtype: object
****************************************
Knowledge sort:
object
****************************************
Worth depend:
Depend  Frequency (%)
Housing_type                             
Home / condo     1107      89.418417
With dad and mom            58       4.684976
Municipal condo     45       3.634895
Rented condo        18       1.453958
Workplace condo         6       0.484653
Co-op condo          4       0.323102pie_chart_plot(cc_train_original,'Housing_type')

bar_plot(cc_train_original, 'Housing_type')

• we are able to see 89.42 % candidates lives in Home/condo

Annual_income

pd.set_option('show.float_format', lambda x: '%.2f' % x)
feature_info(cc_train_original,'Annual_income')Description:
depend      1222.00
imply     194770.84
std      117728.60
min       33750.00
25%      126000.00
50%      172125.00
75%      225000.00
max     1575000.00
Title: Annual_income, dtype: float64
****************************************
Knowledge sort:
float64
****************************************
Worth depend:
Depend  Frequency (%)
Annual_income                      
135000.00        135          11.05
112500.00        114           9.33
180000.00        107           8.76
157500.00         95           7.77
225000.00         91           7.45
...              ...            ...
787500.00          1           0.08
175500.00          1           0.08
95850.00           1           0.08
215100.00          1           0.08
382500.00          1           0.08
[104 rows x 2 columns]box_plot(cc_train_original,'Annual_income')

hist_plot(cc_train_original,'Annual_income')

# bivariate evaluation with goal variable
low_vs_high_risk_box_plot(cc_train_original,'Annual_income')Is_high_risk
0   193525.39
1   204396.43
Title: Annual_income, dtype: float64

• we are able to see common revenue is 194770.84 however this quantity is outlier, Most individuals make 172125, if we ignore the the outliers
• we are able to see within the field plot it’s positively skewed
• low danger and excessive danger candidates nearly related revenue

Type_Occupation

feature_info(cc_train_original, 'Type_Occupation')Description:
depend          838
distinctive          18
prime       Laborers
freq           210
Title: Type_Occupation, dtype: object
****************************************
Knowledge sort:
object
****************************************
Worth depend:
Depend  Frequency (%)
Type_Occupation                            
Laborers                 210          25.06
Core workers               141          16.83
Managers                 112          13.37
Gross sales workers               91          10.86
Drivers                   70           8.35
Excessive ability tech workers     51           6.09
Medication workers            39           4.65
Accountants               37           4.42
Safety workers            18           2.15
Cooking workers             17           2.03
Personal service workers     15           1.79
Cleansing workers            14           1.67
Secretaries                7           0.84
Low-skill Laborers         5           0.60
Waiters/barmen workers       5           0.60
HR workers                   3           0.36
IT workers                   2           0.24
Realty brokers              1           0.12Type_Occupation_nan_count = cc_train_original['Type_Occupation'].isna().sum()
Type_Occupation_nan_count400rows_total_count = cc_train_original.form[0]print('The lacking worth share is {:.2f} %'.format(Type_Occupation_nan_count * 100 / rows_total_count))The lacking worth share is 32.31 %bar_plot(cc_train_original,'Type_Occupation')

• we are able to see most Kind Occupation is Laborers with 210(25.06%), adopted by Core workers with 141(16.83%)
• Even have 32.31 % of lacking knowledge

Type_Income

feature_info(cc_train_original, 'Type_Income')Description:
depend        1238
distinctive          4
prime       Working
freq          634
Title: Type_Income, dtype: object
****************************************
Knowledge sort:
object
****************************************
Worth depend:
Depend  Frequency (%)
Type_Income                               
Working                 634          51.21
Industrial affiliate    292          23.59
Pensioner               216          17.45
State servant            96           7.75bar_plot(cc_train_original,'Type_Income')

pie_chart_plot(cc_train_original,'Type_Income')

• Most candidates are working(51.21%), second is Industrial affiliate(23.59%)

EDUCATION

feature_info(cc_train_original,'EDUCATION')Description:
depend                              1238
distinctive                                5
prime       Secondary / secondary particular
freq                                820
Title: EDUCATION, dtype: object
****************************************
Knowledge sort:
object
****************************************
Worth depend:
Depend  Frequency (%)
EDUCATION                                          
Secondary / secondary particular    820          66.24
Greater training                 349          28.19
Incomplete greater                 54           4.36
Decrease secondary                   14           1.13
Tutorial diploma                    1           0.08pie_chart_plot(cc_train_original,'EDUCATION')

bar_plot(cc_train_original,'EDUCATION')

• Most of candidates have accomplished their Secondary / secondary particular diploma(66.24%)

Employed_days

feature_info(cc_train_original,'Employed_days')Description:
depend   1030.00
imply       7.32
std        6.54
min        0.20
25%        2.57
50%        5.30
75%        9.60
max       40.76
Title: Employed_days, dtype: float64
****************************************
Knowledge sort:int64# beneath employed_days are in years
box_plot(cc_train_original,'Employed_days')

# beneath Employed_days are in 12 months
hist_plot(cc_train_original,'Employed_days')

# bivaraiate evaluation with goal variable, right here 0 means No and 1 means Sure
low_vs_high_risk_box_plot(cc_train_original,'Employed_days')Is_high_risk
0   7.58
1   5.27
Title: Employed_days, dtype: float64

• Right here we are able to see that many of the candidates have been working between 5 to 7 years on common
• Even have many outlier who’ve been working for greater than 20 years.
• Additionally we are able to see employed days[in year] histogram is positively skewed.
• Additionally we are able to see excessive danger have low employment days

Car_Owner

feature_info(cc_train_original,'Car_Owner')Description:
depend     1238
distinctive       2
prime          N
freq       750
Title: Car_Owner, dtype: object
****************************************
Knowledge sort:
object
****************************************
Worth depend:
Depend  Frequency (%)
Car_Owner                      
N            750          60.58
Y            488          39.42bar_plot(cc_train_original,'Car_Owner')

pie_chart_plot(cc_train_original,'Car_Owner')

• we are able to see many of the candidates don’t have a automobile

Propert_Owner

feature_info(cc_train_original,'Propert_Owner')Description:
depend     1238
distinctive       2
prime          Y
freq       805
Title: Propert_Owner, dtype: object
****************************************
Knowledge sort:
object
****************************************
Worth depend:
Depend  Frequency (%)
Propert_Owner                      
Y                805          65.02
N                433          34.98bar_plot(cc_train_original,'Propert_Owner')

pie_chart_plot(cc_train_original,'Propert_Owner')

• we are able to see many of the candidates personal a property(65.02%)

Work_Phone

feature_info(cc_train_original,'Work_Phone')Description:
depend   1238.00
imply       0.22
std        0.41
min        0.00
25%        0.00
50%        0.00
75%        0.00
max        1.00
Title: Work_Phone, dtype: float64
****************************************
Knowledge sort:
int64
****************************************
Worth depend:
Depend  Frequency (%)
Work_Phone                      
0             970          78.35
1             268          21.65bar_plot(cc_train_original,'Work_Phone')

pie_chart_plot(cc_train_original,'Work_Phone')

• Right here we are able to see 78.35% candidates don’t have work cellphone
• Right here 0 means No and 1 means Sure

EMAIL_ID

feature_info(cc_train_original,'EMAIL_ID')Description:
depend   1238.00
imply       0.09
std        0.29
min        0.00
25%        0.00
50%        0.00
75%        0.00
max        1.00
Title: EMAIL_ID, dtype: float64
****************************************
Knowledge sort:
int64
****************************************
Worth depend:
Depend  Frequency (%)
EMAIL_ID                      
0          1123          90.71
1           115           9.29bar_plot(cc_train_original,'EMAIL_ID')

pie_chart_plot(cc_train_original,'EMAIL_ID')

• Right here we are able to see greater than 90% of candidates don’t have e-mail id, solely lower than 10% candidates have e-mail id.
• Right here 0 means No and 1 means sure

Mobile_phone

feature_info(cc_train_original,'Mobile_phone')Description:
depend   1238.00
imply       1.00
std        0.00
min        1.00
25%        1.00
50%        1.00
75%        1.00
max        1.00
Title: Mobile_phone, dtype: float64
****************************************
Knowledge sort:
int64
****************************************
Worth depend:
Depend  Frequency (%)
Mobile_phone                      
1              1238         100.00bar_plot(cc_train_original,'Mobile_phone')

pie_chart_plot(cc_train_original,'Mobile_phone')

• Right here we are able to see all candidates have cell phone

Cellphone

feature_info(cc_train_original,'Cellphone')Description:
depend   1238.00
imply       0.31
std        0.46
min        0.00
25%        0.00
50%        0.00
75%        1.00
max        1.00
Title: Cellphone, dtype: float64
****************************************
Knowledge sort:
int64
****************************************
Worth depend:
Depend  Frequency (%)
Cellphone                      
0        859          69.39
1        379          30.61bar_plot(cc_train_original,'Cellphone')

pie_chart_plot(cc_train_original,'Cellphone')

• Right here we are able to see 69.39% candidates don’t have cellphone,and 30.61% candidates have cellphone.
• Right here 0 means No and 1 means Sure

Is_high_risk(Goal variable)

feature_info(cc_train_original,'Is_high_risk')Description:
depend   1238.00
imply       0.12
std        0.32
min        0.00
25%        0.00
50%        0.00
75%        0.00
max        1.00
Title: Is_high_risk, dtype: float64
****************************************
Knowledge sort:
int64
****************************************
Worth depend:
Depend  Frequency (%)
Is_high_risk                      
0              1093          88.29
1               145          11.71bar_plot(cc_train_original,'Is_high_risk')

pie_chart_plot(cc_train_original,'Is_high_risk')

• Right here we are able to see that 88.29% No danger, it means good candidates ,bank card can be accredited and 11.71% has danger, it means dangerous candidates , bank card is not going to be accredited.
• Right here we are able to see have a imbalanced knowledge that must balanced utilizing SMOTE earlier than coaching our mannequin
• Right here 0 means No and 1 means Sure

3.4.Bivariate evaluation

Numerical vs Numerical Options

scatter plot

sns.pairplot(cc_train_original[cc_train_original['Employed_days'] < 0].drop(['Ind_ID','Mobile_phone','Work_Phone','Phone','EMAIL_ID','Is_high_risk'], axis = 1),nook = True)<seaborn.axisgrid.PairGrid at 0x16a045aeb40>

• “Right here, we observe a constructive linear correlation between the variety of Household Members and the depend of Youngsters. This suggests that as somebody has extra kids, the Household Member depend additionally will increase. Nonetheless, this correlation introduces multicollinearity, an issue arising from two extremely correlated options, which isn’t appropriate for coaching our mannequin. Subsequently, we have to drop these options.
• Moreover, we discover the same sample between Employed Days and Age. The longer the interval of employment, the older somebody tends to be.

Household Member vs CHILDREN(Numerical vs numerical)

sns.regplot(x = 'CHILDREN',y = 'Family_Members', knowledge = cc_train_original, line_kws = {'colour':'inexperienced'})
plt.present()

• Right here, we observe that the extra kids an individual has, the bigger the household measurement

Employed days vs Age(numerical vs numerical)

# Calculate absolutely the values of age and employed days
y_age = np.abs(cc_train_original['Age']) / 365.25
x_employed_days = np.abs(cc_train_original[cc_train_original['Employed_days'] < 0]['Employed_days']) / 365.25
# Create a scatter plot utilizing sns.scatterplot
fig, ax = plt.subplots(figsize=(12, 8))
sns.scatterplot(knowledge=cc_train_original, x=x_employed_days, y=y_age,alpha=0.50)
# Set customized tick values for higher readability
plt.xticks(np.arange(0, x_employed_days.max(), 2.5))
plt.yticks(np.arange(20, y_age.max(), 5))
# Present the plot
plt.present()

• Right here we observe a correlation between the age of candidates and their employed days on this scatter plot.

Heatmap

cc_train_modified = cc_train_original.copy()
# drop Mobile_phone and Ind_ID options
cc_train_modified.drop(['Mobile_phone', 'Ind_ID'], axis=1, inplace=True)
# Choose numeric columns
numeric_df = cc_train_modified.select_dtypes(embrace=['number'])
# Calculate correlation matrix
corr_matrix = numeric_df.corr()
# Create a masks for the higher triangle
masks = np.triu(np.ones_like(corr_matrix, dtype=bool))
# Arrange the matplotlib determine
fig, ax = plt.subplots(figsize=(12, 10))
# Draw the heatmap with the masks and proper side ratio
sns.heatmap(corr_matrix, masks=masks, cmap='coolwarm', vmax=.3, middle=0, annot=True,
sq.=True, linewidths=.5, cbar_kws={"shrink": .5})
# Present the plot
plt.present()

• Right here we observe a excessive correlation between Household Members and Youngsters. Moreover:
• Age reveals some constructive correlation with Household Members and Youngsters, indicating bigger household sizes for older people.
• We will see there isn’t any options that’s correlated with the goal variable.
• There’s a constructive correlation between having a Cellphone and a Work_phone.
• A destructive correlation is obvious between Employed_days and Age.
• Moreover, a constructive correlation exists between Age and Work_phone

Numerical vs Categorical Options(ANOVA[Analysis of Variance])

Age vs categorical options

fig, axes = plt.subplots(4,2,figsize=(15,20),dpi=180)
fig.tight_layout(pad=5.0)
cat_features = ['GENDER','Car_Owner','Propert_Owner','Type_Income','EDUCATION','Marital_status','Housing_type','Type_Occupation']
for cat_feature_count, ax in enumerate(axes):
for row_count in vary(4):
for feat_count in vary(2):
sns.boxplot(ax=axes[row_count,feat_count],x=cc_train_original[cat_features[cat_feature_count]],y=np.abs(cc_train_original['Age'])/365.25)
axes[row_count,feat_count].set_title(cat_features[cat_feature_count] + " vs age")
plt.sca(axes[row_count,feat_count])
plt.xticks(rotation=45,ha='proper')
plt.ylabel('Age')
cat_feature_count += 1
break

• Feminine candidates are older than male candidates.
• People and not using a automobile are usually older.
• Property house owners are usually older than these with out property.
• Pensioners are older than those that are at the moment working (with some outliers the place people pensioned at a younger age).
• Widows are usually a lot older, with some outliers of their 30s.
• People dwelling with dad and mom are youthful, with some outliers.
• Safety workers members are usually older, whereas these in data expertise (IT) are usually youthful.

Categorical vs Categorical Options(Chi-Sq. check)

• Null Speculation: The classes of the options haven’t any impact on the goal variable.

Various Speculation: A number of characteristic classes have a big impact on the goal variable.

import scipy.stats as stats
def chi_square_test(characteristic):
# Choose rows with excessive danger
high_risk_ft = cc_train_original[cc_train_original['Is_high_risk'] == 1][feature]
high_risk_ft_ct = pd.crosstab(index=high_risk_ft, columns=['Count']).rename_axis(None, axis=1)
# Drop the index characteristic identify
high_risk_ft_ct.index.identify = None
# Observe values
obs = high_risk_ft_ct
print('Noticed values:n')
print(obs)
print('n')
# Anticipated values
exp = pd.DataFrame([obs['Count'].sum() / len(obs)] * len(obs.index), columns=['Count'], index=obs.index)
print('Anticipated values:n')
print(exp)
print('n')
# Chi-square check
chi_squared_stat = (((obs - exp) ** 2) / exp).sum().values[0]
print('Chi-square:n')
print(chi_squared_stat)
print('n')
# Crucial worth
crit = stats.chi2.ppf(q=0.95, df=len(obs) - 1)
print('Crucial worth:n')
print(crit)
print('n')
# P-value
p_value = 1 - stats.chi2.cdf(x=chi_squared_stat, df=len(obs) - 1)
print('P-value:n')
print(p_value)
print('n')
if chi_squared_stat >= crit:
print('Reject the null speculation')
elif chi_squared_stat <= crit:
print('Fail to reject the null speculation')cat_features = ['GENDER','Car_Owner','Propert_Owner','Type_Income','EDUCATION','Marital_status','Housing_type','Type_Occupation']
for feat in cat_features:
print('nn**** {} ****n'.format(feat))
chi_square_test(feat)**** GENDER ****
Noticed values:
Depend
F     77
M     63
Anticipated values:
Depend
F  70.00
M  70.00
Chi-square:
1.4
Crucial worth:
3.841458820694124
P-value:
0.2367235706378572
Fail to reject the null speculation
**** Car_Owner ****
Noticed values:
Depend
N     85
Y     60
Anticipated values:
Depend
N  72.50
Y  72.50
Chi-square:
4.310344827586207
Crucial worth:
3.841458820694124
P-value:
0.0378812823153869
Reject the null speculation
**** Propert_Owner ****
Noticed values:
Depend
N     52
Y     93
Anticipated values:
Depend
N  72.50
Y  72.50
Chi-square:
11.593103448275862
Crucial worth:
3.841458820694124
P-value:
0.0006619684903875767
Reject the null speculation
**** Type_Income ****
Noticed values:
Depend
Industrial affiliate     42
Pensioner                34
State servant             6
Working                  63
Anticipated values:
Depend
Industrial affiliate  36.25
Pensioner             36.25
State servant         36.25
Working               36.25
Chi-square:
46.03448275862069
Crucial worth:
7.814727903251179
P-value:
5.576543671281797e-10
Reject the null speculation
**** EDUCATION ****
Noticed values:
Depend
Greater training                  50
Incomplete greater                  4
Decrease secondary                    5
Secondary / secondary particular     86
Anticipated values:
Depend
Greater training               36.25
Incomplete greater              36.25
Decrease secondary                36.25
Secondary / secondary particular  36.25
Chi-square:
129.12413793103448
Crucial worth:
7.814727903251179
P-value:
0.0
Reject the null speculation
**** Marital_status ****
Noticed values:
Depend
Civil marriage            4
Married                  99
Separated                11
Single / not married     24
Widow                     7
Anticipated values:
Depend
Civil marriage        29.00
Married               29.00
Separated             29.00
Single / not married  29.00
Widow                 29.00
Chi-square:
219.24137931034483
Crucial worth:
9.487729036781154
P-value:
0.0
Reject the null speculation
**** Housing_type ****
Noticed values:
Depend
Co-op condo          1
Home / condo      121
Municipal condo     14
Workplace condo         1
Rented condo         3
With dad and mom             5
Anticipated values:
Depend
Co-op condo      24.17
Home / condo    24.17
Municipal condo  24.17
Workplace condo     24.17
Rented condo     24.17
With dad and mom         24.17
Chi-square:
470.43448275862056
Crucial worth:
11.070497693516351
P-value:
0.0
Reject the null speculation
**** Type_Occupation ****
Noticed values:
Depend
Accountants                5
Cleansing workers             1
Cooking workers              4
Core workers                21
Drivers                    7
Excessive ability tech workers      5
IT workers                   2
Laborers                  22
Low-skill Laborers         1
Managers                  11
Medication workers             3
Gross sales workers                8
Safety workers             7
Waiters/barmen workers       1
Anticipated values:
Depend
Accountants             7.00
Cleansing workers          7.00
Cooking workers           7.00
Core workers              7.00
Drivers                 7.00
Excessive ability tech workers   7.00
IT workers                7.00
Laborers                7.00
Low-skill Laborers      7.00
Managers                7.00
Medication workers          7.00
Gross sales workers             7.00
Safety workers          7.00
Waiters/barmen workers    7.00
Chi-square:
86.28571428571428
Crucial worth:
22.362032494826934
P-value:
7.139844271364382e-13
Reject the null speculation

Abstract of EDA

Gender Distribution

• Two distinctive lessons, Feminine (F) and Male (M), with 778 and 454 candidates, respectively.
• 63.14% are feminine, and 36.85% are male.

Marital Standing

• 5 distinctive lessons, with most candidates (68.50%) being married.
• Widows have the next danger in comparison with separated people.

Household Members

• Numerical characteristic with a median of two relations.
• Most candidates have 2 relations (51.85%).

Youngsters

• Most candidates don’t have kids.
• Three outliers with 5, 6, and 15 kids.

Housing Kind

• 89.42% of candidates dwell in a home/condo.

Annual Revenue

• Common revenue is 194,770.84, with outliers.
• Most individuals make 172,125, ignoring outliers.
• Positively skewed distribution.

Kind Occupation

• Laborers (25.06%) and Core workers (16.83%) are the commonest occupations.
• 32.31% lacking knowledge.

Kind Revenue

• Most candidates are working (51.21%), adopted by industrial associates (23.59%).

Schooling

• Most candidates accomplished Secondary/Secondary Particular (66.24%).

Employed Days

• Most candidates have been working between 5 to 7 years on common.
• Outliers with employment durations over 20 years.

Automobile Possession and Property Possession

• Most candidates don’t have a automobile (70.36%).
• Most candidates personal property (65.02%).

Cellphone and Work Cellphone

• 78.35% of candidates don’t have a piece cellphone.
• Greater than 90% don’t have an e-mail ID.
• All candidates have a cell phone.
• 69.39% don’t have a cellphone, and 30.61% have a cellphone.

Is Excessive Danger (Goal Variable)

• 88.29% haven’t any danger, and 11.71% have danger, indicating imbalanced knowledge.
• Imbalance must be addressed utilizing SMOTE earlier than mannequin coaching.

Age and Is Excessive Danger

• There isn’t any important distinction in age between high-risk and low-risk candidates.
• The imply age for each teams is round 43 years, and there’s no correlation between age and danger elements.

Correlations

• Constructive linear correlation between Household Members and Youngsters, introducing multicollinearity.
• Related sample noticed between Employed Days and Age, indicating longer employment correlates with older age.
• No options are strongly correlated with the goal variable.
• Constructive correlation between having a Cellphone and a Work_phone.
• Unfavorable correlation between Employed_days and Age.
• Constructive correlation between Age and Work_phone.

Demographic Observations

• Feminine candidates are older than male candidates.
• Non-car house owners are usually older.
• Property house owners are usually older.
• Pensioners are older than working people, with some outliers.
• Widows are usually older.
• People dwelling with dad and mom are youthful, with some outliers.
• Safety workers tends to be older, whereas IT workers tends to be youthful.

4.1.Here’s a listing of all of the transformations that must be utilized to every characteristic

Ind_ID

GENDER

• One-hot Encoding
• impute lacking worth

Car_Owner

• Change it numerical
• One-hot encoding

Propert_Owner

• Change it numerical
• One-hot encoding

CHILDREN

Annual_income

• Take away outlier
• Repair Skewness
• Min-Max Scaling
• impute lacking worth

Type_Income

EDUCATION

Marital_status

Housing_type

Age

• Min-Max Scaling
• Repair Skewness
• Abs worth and divide 365.25
• impute lacking worth

Employed_days

• Min-Max Scaling
• Take away outlier
• Abs worth and divide 365.25

Mobile_phone

Work_Phone

Cellphone

EMAIL_ID

Type_Occupation

• One-hot encoding
• impute lacking worth

Family_Members

Is_high_risk

• Stability the Imbalance Knowledge with SMOTE

4.2.Knowledge Cleansing

4.3.1.Impute Lacking Worth

class Handle_Missing_Values(BaseEstimator, TransformerMixin):
def __init__(self, categorical_cols=['GENDER', 'Type_Occupation'], numerical_cols=['Annual_income', 'Age']):
self.categorical_cols = categorical_cols
self.numerical_cols = numerical_cols
def match(self, df):
# Retailer the mode for categorical columns and the imply for numerical columns
self.imputation_values = {}
for col in self.categorical_cols:
self.imputation_values[col] = df[col].mode()[0]
for col in self.numerical_cols:
self.imputation_values[col] = df[col].imply()
return self
def remodel(self, df):
# Change lacking values with the saved imputation values
for col in self.categorical_cols:
df[col].fillna(self.imputation_values[col], inplace=True)
for col in self.numerical_cols:
df[col].fillna(self.imputation_values[col], inplace=True)
return df#cc_train_original.isnull().sum()

4.3.2.Outlier Remover Perform

• Right here we’re creating a category to deal with outliers.
• This class will take away outliers roughly than 3 inter-quantile ranges away from the imply

class OutlierRemover(BaseEstimator, TransformerMixin):
def __init__(self,feat_with_outliers = ['Family_Members','Annual_income', 'Employed_days']):
self.feat_with_outliers = feat_with_outliers
def match(self,df):
return self
def remodel(self,df):
if (set(self.feat_with_outliers).issubset(df.columns)):
# 25% quantile
Q1 = df[self.feat_with_outliers].quantile(.25)
# 75% quantile
Q3 = df[self.feat_with_outliers].quantile(.75)
IQR = Q3 - Q1
# preserve the info inside 3 IQR
df = df[~((df[self.feat_with_outliers] < (Q1 - 3 * IQR)) |(df[self.feat_with_outliers] > (Q3 + 3 * IQR))).any(axis=1)]
return df
else:
print("Warning: A number of specified options for outlier elimination will not be current within the dataframe. No outliers eliminated.")
return df

4.3.Characteristic Choice