KPMG Data Analytics Virtual Internship

Task 2: Data Insights

Contents

  • Data Exploration
  • Model Development
  • Interpretation
In [1]:
import pandas as pd
import numpy as np
import seaborn as sns

import matplotlib.pyplot as plt
import matplotlib.ticker as mtick
%matplotlib inline

from imblearn.over_sampling import SMOTE

from sklearn.ensemble import RandomForestClassifier
from sklearn.metrics import (accuracy_score, classification_report, f1_score,
                             precision_recall_curve, precision_score,
                             recall_score, roc_auc_score, roc_curve, RocCurveDisplay)
from sklearn.model_selection import (GridSearchCV, KFold, RandomizedSearchCV,
                                     ShuffleSplit, StratifiedKFold,
                                     StratifiedShuffleSplit, cross_val_predict,
                                     cross_val_score, learning_curve,
                                     train_test_split)
from sklearn.preprocessing import LabelEncoder
from sklearn.tree import DecisionTreeClassifier

pd.set_option('display.max_columns', 50)

import warnings
warnings.filterwarnings("ignore")
In [2]:
df2 = pd.read_excel('KPMG_task1.xlsx', sheet_name = [0,1,2,3])
Transactions = df2[0]
NewCustomerList = df2[1]
CustomerDemographic = df2[2]
CustomerAddress = df2[3]

1. Data Exploration¶

1.1. Gender distributions¶

In [3]:
fig, ax = plt.subplots(1, 2, figsize=(15, 5))
palette_color = sns.color_palette('RdBu_r')
ax[0].pie(CustomerDemographic['gender'].value_counts(), 
          labels=CustomerDemographic['gender'].value_counts().index, colors=palette_color,
          autopct='%1.1f%%', wedgeprops = { 'linewidth' : 3, 'edgecolor' : 'white' })
ax[0].set_title('OLD gender distributions', fontsize = 14)

ax[1].pie(NewCustomerList['gender'].value_counts(), 
          labels=NewCustomerList['gender'].value_counts().index, colors=palette_color,
          autopct='%1.1f%%', wedgeprops = { 'linewidth' : 3, 'edgecolor' : 'white' })
ax[1].set_title('NEW gender distributions', fontsize = 14)
plt.show()
  • The graphs illustrate the proportion of male and female customers in old and new datasets.
  • Specifically, in both existing and new 1000 customers data, about half of the customers are male while the other half are female.
  • The number of females is slightly higher than that of males.

1.2 Age distributions¶

In [3]:
#create "age" column
CustomerDemographic['age'] =  2022 - CustomerDemographic['DOB'].dt.year
CustomerDemographic.drop(['DOB'], inplace = True, axis = 1)
NewCustomerList['age'] =  2022 - NewCustomerList['DOB'].dt.year
NewCustomerList.drop(['DOB'], inplace = True, axis = 1)
In [5]:
fig, ax = plt.subplots(1, 2, figsize=(15, 5))
sns.histplot(data = CustomerDemographic['age'], color = '#2f528f', ax = ax[0])
ax[0].set_ylabel('# Customers', fontsize = 13)
ax[0].set_xlabel('Age', fontsize = 13)
ax[0].set_title('Number of OLD customers by age', fontsize = 14)
sns.histplot(data = NewCustomerList['age'], color = '#2f528f', ax = ax[1]) 
ax[1].set_ylabel('# Customers', fontsize = 13)
ax[1].set_xlabel('Age', fontsize = 13)
ax[1].set_title('Number of NEW customers by age', fontsize = 14)
plt.show()

As we can see from both graphs, the highest number of customers is in the age between 40 and 50 years old.

In [4]:
#create "age_group" column
def ages(x):
    if (20 <= x <= 39) :
        return 'Adults'
    elif (40 <= x <= 59):
        return 'Middle-Aged People'
    elif (60 <= x): 
        return 'The Elderly'
CustomerDemographic['age_group'] = CustomerDemographic['age'].apply(lambda x: ages(x))
NewCustomerList['age_group'] = NewCustomerList['age'].apply(lambda x: ages(x))
In [7]:
fig, ax = plt.subplots(1, 2, figsize=(15, 5))
sns.countplot(x='age_group', data = CustomerDemographic, color='#2f528f', ax = ax[0])
ax[0].bar_label(ax[0].containers[0])
ax[0].set_ylabel('# Customers', fontsize = 13)
ax[0].set_xlabel('Age Group', fontsize = 13)
ax[0].set_title('Number of OLD customers by age group', fontsize = 14)
sns.countplot(x='age_group', data = NewCustomerList, color='#2f528f', ax = ax[1])
ax[1].bar_label(ax[1].containers[0])
ax[1].set_ylabel('# Customers', fontsize = 13)
ax[1].set_xlabel('Age Group', fontsize = 13)
ax[1].set_title('Number of NEW customers by age group', fontsize = 14)
plt.show()
  • The group of middle-aged people also makes up the highest proportion in 3 age groups.
  • However, there is an increase in the number of the elderly in term of percentage in the new dataset.

1.3. Bike related purchases over the last 3 years¶

In [8]:
print('5 products that have highest number of purchases by OLD customers in the last 3 years:'
      , CustomerDemographic['past_3_years_bike_related_purchases'].value_counts().head(5).to_dict())

5 products that have highest number of purchases by OLD customers in the last 3 years: {19: 55, 16: 55, 20: 53, 67: 52, 2: 50}
In [9]:
print('5 products that have highest number of purchases by NEW customers in the last 3 years:'
      , NewCustomerList['past_3_years_bike_related_purchases'].value_counts().head(5).to_dict())

5 products that have highest number of purchases by NEW customers in the last 3 years: {60: 20, 59: 18, 42: 17, 70: 17, 37: 16}

1.4. Job industry category¶

In [10]:
fig, ax = plt.subplots(1, 2, figsize=(15, 5))
sns.countplot(x='job_industry_category', data = CustomerDemographic, color='#2f528f', ax = ax[0])
ax[0].bar_label(ax[0].containers[0])
ax[0].set_ylabel('# Customers', fontsize = 13)
ax[0].set_xlabel('Job industry category', fontsize = 13)
ax[0].set_title('Number of OLD customers by job industry category', fontsize = 14)
sns.countplot(x='job_industry_category', data = NewCustomerList, color='#2f528f', ax = ax[1])
ax[1].bar_label(ax[1].containers[0])
ax[1].set_ylabel('# Customers', fontsize = 13)
ax[1].set_xlabel('Job industry category', fontsize = 13)
ax[1].set_title('Number of NEW customers by job industry category', fontsize = 14)
ax[0].tick_params(labelrotation=90)
ax[1].tick_params(labelrotation=90)
plt.show()
  • The majority of our customers are working at Manufacturing and Finance industries.
  • The order of job industries ranked by number of customers are the same between existing dataset and new one.

1.5. Wealth segments¶

In [11]:
fig, ax = plt.subplots(1, 2, figsize=(15, 5))
sns.histplot(data = CustomerDemographic, x = 'wealth_segment', hue = 'age_group', discrete=True
             , multiple="stack", palette = 'RdBu_r', shrink=.8, ax = ax[0])
ax[0].set_ylabel('# Customers', fontsize = 13)
ax[0].set_xlabel('Wealth segments', fontsize = 13)
ax[0].set_title('OLD Wealth segments by age group', fontsize = 14)
for c in ax[0].containers:
    labels = [f'{h/CustomerDemographic.age_group.count()*100:0.1f}%' if (h := v.get_height()) > 0 else '' for v in c]
    ax[0].bar_label(c, labels=labels, label_type='edge')
sns.histplot(data = NewCustomerList, x = 'wealth_segment', hue = 'age_group', discrete=True
             , multiple="stack", palette = 'RdBu_r', shrink=.8, ax = ax[1])
ax[1].set_ylabel('# Customers', fontsize = 13)
ax[1].set_xlabel('Wealth segments', fontsize = 13)
ax[1].set_title('NEW Wealth segments by age group', fontsize = 14)
for c in ax[1].containers:
    labels = [f'{h/NewCustomerList.age_group.count()*100:0.1f}%' if (h := v.get_height()) > 0 else '' for v in c]
    ax[1].bar_label(c, labels=labels, label_type='edge')
plt.show()
  • In all age groups, the number of Mass Customers is the highest, so we should focus on this social class.
  • Affluent Customer and High Net Worth have quite similar proportions of customers.
  • In the new dataset, the percentages of elderly people rise by 2 times in all wealth segments.

1.6. Cars owned on each states¶

In [12]:
old_dataset = CustomerDemographic.merge(CustomerAddress, how='left', on='customer_id')
In [13]:
fig, ax = plt.subplots(1, 2, figsize=(15, 5))
sns.histplot(data = old_dataset, x = 'state', hue = 'owns_car', discrete=True
             , multiple="stack", palette = 'RdBu_r', shrink=.8, ax = ax[0])
ax[0].set_ylabel('# Customers', fontsize = 13)
ax[0].set_xlabel('States', fontsize = 13)
ax[0].set_title('OLD cars owned or not in each state', fontsize = 14)
for c in ax[0].containers:
    labels = [f'{h/old_dataset.owns_car.count()*100:0.1f}%' if (h := v.get_height()) > 0 else '' for v in c]
    ax[0].bar_label(c, labels=labels, label_type='edge')
sns.histplot(data = NewCustomerList, x = 'state', hue = 'owns_car', discrete=True
             , multiple="stack", palette = 'RdBu_r', shrink=.8, ax = ax[1])
ax[1].set_ylabel('# Customers', fontsize = 13)
ax[1].set_xlabel('States', fontsize = 13)
ax[1].set_title('NEW cars owned or not in each state', fontsize = 14)
for c in ax[1].containers:
    labels = [f'{h/NewCustomerList.owns_car.count()*100:0.1f}%' if (h := v.get_height()) > 0 else '' for v in c]
    ax[1].bar_label(c, labels=labels, label_type='edge')
plt.show()
  • The number of customers is highest in New South Wales, followed by Victoria and Queensland respectively.
  • It can be seen that about half of the customers have cars, while the other half do not.
  • However, there are fewer New South Wales people owning cars in the new 1000 customers dataset than the old one.

1.7. RFM analysis and customer classifications¶

In [5]:
#create "profit" column
Transactions['profit'] = Transactions['list_price'] - Transactions['standard_cost']
In [6]:
#create "recency" column
Transactions['recency'] = Transactions['transaction_date'].apply(lambda x: (Transactions['transaction_date'].max() - x).days)
In [7]:
#create "rfm" table
aggr = {
    'recency': lambda x: x.min(),  # the number of days since last order (Recency)
    'product_id': lambda x: x.count(), # the total number of products bought in the last period (Frequency)
    'profit': lambda x: x.sum()
}
rfm = Transactions.groupby('customer_id').agg(aggr).reset_index()
rfm.rename(columns={'product_id': 'frequency', 'profit': 'monetary'}, inplace=True)
rfm.head()
Out[7]:
customer_id recency frequency monetary
0 1 7 11 3018.09
1 2 128 3 2226.26
2 3 102 8 3362.81
3 4 195 2 220.57
4 5 16 6 2394.94
In [8]:
quartiles = rfm[['recency', 'frequency', 'monetary']].quantile([.25, .5, .75]).to_dict()
quartiles
Out[8]:
{'recency': {0.25: 17.0, 0.5: 44.0, 0.75: 86.0},
 'frequency': {0.25: 4.0, 0.5: 6.0, 0.75: 7.0},
 'monetary': {0.25: 1841.73, 0.5: 2862.330000000001, 0.75: 4183.81}}
In [9]:
def r_score(x):
    if x <= quartiles['recency'][.25]:
        return 4
    elif x <= quartiles['recency'][.5]:
        return 3
    elif x <= quartiles['recency'][.75]:
        return 2
    else:
        return 1

def fm_score(x, c):
    if x <= quartiles[c][.25]:
        return 1
    elif x <= quartiles[c][.5]:
        return 2
    elif x <= quartiles[c][.75]:
        return 3
    else:
        return 4
In [10]:
rfm['R'] = rfm['recency'].apply(lambda x: r_score(x))
rfm['F'] = rfm['frequency'].apply(lambda x: fm_score(x, 'frequency'))
rfm['M'] = rfm['monetary'].apply(lambda x: fm_score(x, 'monetary'))
In [11]:
rfm['RFM Score'] = rfm['R'].map(str) + rfm['F'].map(str) + rfm['M'].map(str)
rfm['RFM Score'] = pd.to_numeric(rfm['RFM Score'], errors='coerce')
rfm.head()
Out[11]:
customer_id recency frequency monetary R F M RFM Score
0 1 7 11 3018.09 4 4 3 443
1 2 128 3 2226.26 1 1 2 112
2 3 102 8 3362.81 1 4 3 143
3 4 195 2 220.57 1 1 1 111
4 5 16 6 2394.94 4 2 2 422
In [12]:
quartiles_segment = rfm[['RFM Score']].quantile([.25, .5, .75]).to_dict()
quartiles_segment
Out[12]:
{'RFM Score': {0.25: 211.0, 0.5: 311.0, 0.75: 411.0}}
In [13]:
def segment(x):
    if x <= quartiles_segment['RFM Score'][.25]:
        return 'Bronze Customer'
    elif x <= quartiles_segment['RFM Score'][.5]:
        return 'Silver Customer'
    elif x <= quartiles_segment['RFM Score'][.75]:
        return 'Gold Customer'
    else:
        return 'Platinum Customer'
In [14]:
rfm['Segment'] = rfm['RFM Score'].apply(lambda x: segment(x))
rfm.head()
Out[14]:
customer_id recency frequency monetary R F M RFM Score Segment
0 1 7 11 3018.09 4 4 3 443 Platinum Customer
1 2 128 3 2226.26 1 1 2 112 Bronze Customer
2 3 102 8 3362.81 1 4 3 143 Bronze Customer
3 4 195 2 220.57 1 1 1 111 Bronze Customer
4 5 16 6 2394.94 4 2 2 422 Platinum Customer
In [24]:
ax = (rfm['Segment'].value_counts()*100.0 /len(rfm)).plot(kind='bar', stacked = True, rot = 0, color = '#2f528f', figsize = (7,5))

ax.yaxis.set_major_formatter(mtick.PercentFormatter())
ax.set_ylabel('% Customers', fontsize = 13)
ax.set_xlabel('Customer Group', fontsize = 13)
ax.set_title('Customer Classification Rate', fontsize = 14)

totals = []
for i in ax.patches:
    totals.append(i.get_width())
total = sum(totals)
for i in ax.patches:
    # get_width pulls left or right; get_y pushes up or down
    ax.text(i.get_x()+.05, i.get_height()-2.0, \
            str(round((i.get_height()/total), 1))+'%', fontsize=12, color='white', weight = 'bold')
  • We segment customers who have already had prior transactions into 4 groups: Platinum, Gold, Silver and Gold Customer.
  • Platinum customers will drive the highest profit to the company, however, this group marks up the lowest percent.
  • We will consider Platinum and Gold customers as potential customers to be targeted in the next marketing campaign and use this result to train model in the next part.

2. Model Development¶

Data preprosessing¶

In [25]:
old_dataset = old_dataset.merge(rfm, how='left', on='customer_id')
In [26]:
#create "target" column
def target(x):
    if (x == 'Platinum Customer') | (x == 'Gold Customer'):
        return 1
    else:
        return 0
old_dataset['target'] = old_dataset['Segment'].apply(lambda x: target(x))
In [27]:
#drop some unneccessary columns
old_dataset.drop(['customer_id', 'recency', 'frequency', 'monetary', 'R', 'F', 'M', 'RFM Score', 'Segment',
                 'deceased_indicator', 'age_group', 'country', 'name', 'address'], inplace = True, axis = 1)
old_dataset.head()
Out[27]:
gender past_3_years_bike_related_purchases job_title job_industry_category wealth_segment owns_car tenure age postcode state property_valuation target
0 Female 93 Executive Secretary Health Mass Customer Yes 11 69 2016.0 NSW 10.0 1
1 Male 81 Administrative Officer Financial Services Mass Customer Yes 16 42 2153.0 NSW 10.0 0
2 Male 61 Recruiting Manager Property Mass Customer Yes 15 68 NaN NaN NaN 0
3 Male 33 Recruiting Manager IT Mass Customer No 7 61 4211.0 QLD 9.0 0
4 Female 56 Senior Editor IT Affluent Customer Yes 8 45 2448.0 NSW 4.0 1
In [28]:
#remove missing values
old_dataset.dropna(inplace=True)
old_dataset.reset_index(drop=True, inplace=True)
In [29]:
#transform categorical variables into columns with numerical values by label encoding
le = LabelEncoder()
old_dataset['gender'] = le.fit_transform(old_dataset['gender'])
old_dataset['job_title'] = le.fit_transform(old_dataset['job_title'])
old_dataset['job_industry_category'] = le.fit_transform(old_dataset['job_industry_category'])
old_dataset['wealth_segment'] = le.fit_transform(old_dataset['wealth_segment'])
old_dataset['owns_car'] = le.fit_transform(old_dataset['owns_car'])
old_dataset['state'] = le.fit_transform(old_dataset['state'])
old_dataset
Out[29]:
gender past_3_years_bike_related_purchases job_title job_industry_category wealth_segment owns_car tenure age postcode state property_valuation target
0 0 93 71 3 2 1 11 69 2016.0 0 10.0 1
1 1 81 19 2 2 1 16 42 2153.0 0 10.0 0
2 1 33 133 4 2 0 7 61 4211.0 1 9.0 0
3 0 56 149 4 0 1 8 45 2448.0 0 4.0 1
4 1 35 149 7 1 1 13 56 3216.0 2 9.0 0
... ... ... ... ... ... ... ... ... ... ... ... ...
3901 1 93 168 5 2 1 14 47 2088.0 0 12.0 0
3902 0 8 184 3 2 0 19 47 3977.0 2 6.0 0
3903 0 87 170 5 1 1 1 21 2350.0 0 2.0 0
3904 1 11 170 5 0 1 10 49 3064.0 2 3.0 0
3905 1 76 158 5 0 0 11 31 4511.0 1 6.0 0

3906 rows × 12 columns

In [30]:
#check data skewness
target_1 = old_dataset[old_dataset["target"] == 1]
target_0  = old_dataset[old_dataset["target"] == 0]
print(target_1.shape)
print(target_0.shape)

(1610, 12)
(2296, 12)
In [31]:
X = old_dataset.drop('target', axis=1)
Y = old_dataset['target']
In [32]:
#splitting the original dataset into the Training set and Test set
original_X_train, original_X_test, original_Y_train, original_Y_test = train_test_split(X, Y, test_size = 0.2, random_state=1, stratify = Y)
In [33]:
#random undersampling
# Lets shuffle the data before creating the subsamples
old_dataset = old_dataset.sample(frac=1)

# amount of fraud classes 1610 rows.
target_1 = old_dataset.loc[old_dataset['target'] == 1]
target_0 = old_dataset.loc[old_dataset['target'] == 0][:1610]

normal_distributed_df = pd.concat([target_1, target_0])

# Shuffle dataframe rows
old_df = normal_distributed_df.sample(frac=1, random_state=1)

old_df['target'].value_counts()
Out[33]:
1    1610
0    1610
Name: target, dtype: int64
In [34]:
#splitting the undersampled dataset into the Training set and Test set
X_under = old_df.drop('target', axis=1)
Y_under = old_df['target']
X_train, X_test, Y_train, Y_test = train_test_split(X_under, Y_under, test_size=0.2, random_state=1)

2.1. Random Forest Classifier¶

In [35]:
rf = RandomForestClassifier(class_weight = 'balanced', random_state =22)

rf.fit(X_train, Y_train)

rf_Y_pred = rf.predict(original_X_test)
rf_Y_proba = rf.predict_proba(original_X_test)
In [36]:
print(classification_report(original_Y_test, rf_Y_pred))

              precision    recall  f1-score   support

           0       0.93      0.80      0.86       460
           1       0.76      0.91      0.83       322

    accuracy                           0.84       782
   macro avg       0.84      0.85      0.84       782
weighted avg       0.86      0.84      0.84       782

2.2. Decision Tree Classifier¶

In [37]:
dt = DecisionTreeClassifier(random_state =22)

dt.fit(X_train, Y_train)

dt_Y_pred = dt.predict(original_X_test)
dt_Y_proba = dt.predict_proba(original_X_test)
In [38]:
print(classification_report(original_Y_test, dt_Y_pred))

              precision    recall  f1-score   support

           0       0.93      0.77      0.84       460
           1       0.74      0.92      0.82       322

    accuracy                           0.83       782
   macro avg       0.83      0.85      0.83       782
weighted avg       0.85      0.83      0.83       782

2.3. Model Tuning¶

Random Forest Tuned¶

In [39]:
CV = StratifiedKFold(n_splits=5, shuffle=True, random_state=22)
RF_tuned = RandomForestClassifier()
grid_RF_params = {'bootstrap': [True],
                  'max_depth': [80, 90, 100, 110],
                  'max_features': [2, 3],
                  'min_samples_leaf': [3, 4, 5],
                  'min_samples_split': [8, 10, 12],
                  'n_estimators': [100, 200, 300, 1000],}

grid_RF_cv = GridSearchCV(RF_tuned, grid_RF_params, scoring = 'f1', refit = True, cv=CV, verbose=1)  
grid_RF_cv.fit(X_train, Y_train) 

Fitting 5 folds for each of 288 candidates, totalling 1440 fits
Out[39]:
GridSearchCV(cv=StratifiedKFold(n_splits=5, random_state=22, shuffle=True),
             estimator=RandomForestClassifier(),
             param_grid={'bootstrap': [True], 'max_depth': [80, 90, 100, 110],
                         'max_features': [2, 3], 'min_samples_leaf': [3, 4, 5],
                         'min_samples_split': [8, 10, 12],
                         'n_estimators': [100, 200, 300, 1000]},
             scoring='f1', verbose=1)
In a Jupyter environment, please rerun this cell to show the HTML representation or trust the notebook.
On GitHub, the HTML representation is unable to render, please try loading this page with nbviewer.org.
GridSearchCV(cv=StratifiedKFold(n_splits=5, random_state=22, shuffle=True),
             estimator=RandomForestClassifier(),
             param_grid={'bootstrap': [True], 'max_depth': [80, 90, 100, 110],
                         'max_features': [2, 3], 'min_samples_leaf': [3, 4, 5],
                         'min_samples_split': [8, 10, 12],
                         'n_estimators': [100, 200, 300, 1000]},
             scoring='f1', verbose=1)
RandomForestClassifier()
RandomForestClassifier()
In [40]:
best_RF = grid_RF_cv.best_estimator_
best_RF.fit(X_train, Y_train)
best_rf_Y_pred = best_RF.predict(original_X_test)
best_rf_Y_proba = best_RF.predict_proba(original_X_test)
In [41]:
print(classification_report(original_Y_test, best_rf_Y_pred))

              precision    recall  f1-score   support

           0       0.92      0.80      0.85       460
           1       0.75      0.90      0.82       322

    accuracy                           0.84       782
   macro avg       0.84      0.85      0.84       782
weighted avg       0.85      0.84      0.84       782

Decision Tree Tuned¶

In [42]:
DT_tuned = DecisionTreeClassifier()
grid_DT_params = {'criterion': ['gini', 'entropy', 'log_loss'], 
                  'splitter' : ['best', 'random'], 
                  'max_depth' : [1,3,5,7,9,11,12],
                  'min_samples_leaf' :[1,2,3,4,5,6,7,8,9,10],
                  'max_features' : ['auto', 'sqrt', 'log2', None],}

grid_DT_cv = GridSearchCV(DT_tuned, grid_DT_params, scoring = 'f1', refit = True, cv=CV, verbose=1)  
grid_DT_cv.fit(X_train, Y_train) 

Fitting 5 folds for each of 1680 candidates, totalling 8400 fits
Out[42]:
GridSearchCV(cv=StratifiedKFold(n_splits=5, random_state=22, shuffle=True),
             estimator=DecisionTreeClassifier(),
             param_grid={'criterion': ['gini', 'entropy', 'log_loss'],
                         'max_depth': [1, 3, 5, 7, 9, 11, 12],
                         'max_features': ['auto', 'sqrt', 'log2', None],
                         'min_samples_leaf': [1, 2, 3, 4, 5, 6, 7, 8, 9, 10],
                         'splitter': ['best', 'random']},
             scoring='f1', verbose=1)
In a Jupyter environment, please rerun this cell to show the HTML representation or trust the notebook.
On GitHub, the HTML representation is unable to render, please try loading this page with nbviewer.org.
GridSearchCV(cv=StratifiedKFold(n_splits=5, random_state=22, shuffle=True),
             estimator=DecisionTreeClassifier(),
             param_grid={'criterion': ['gini', 'entropy', 'log_loss'],
                         'max_depth': [1, 3, 5, 7, 9, 11, 12],
                         'max_features': ['auto', 'sqrt', 'log2', None],
                         'min_samples_leaf': [1, 2, 3, 4, 5, 6, 7, 8, 9, 10],
                         'splitter': ['best', 'random']},
             scoring='f1', verbose=1)
DecisionTreeClassifier()
DecisionTreeClassifier()
In [43]:
best_DT = grid_DT_cv.best_estimator_
best_DT.fit(X_train, Y_train)
best_dt_Y_pred = best_DT.predict(original_X_test)
best_dt_Y_proba = best_DT.predict_proba(original_X_test)
In [44]:
print(classification_report(original_Y_test, best_dt_Y_pred))

              precision    recall  f1-score   support

           0       0.60      0.51      0.55       460
           1       0.43      0.52      0.47       322

    accuracy                           0.52       782
   macro avg       0.52      0.52      0.51       782
weighted avg       0.53      0.52      0.52       782

3. Interpretation¶

3.1. Prepare NewCustomerList data¶

In [45]:
NewCustomerList.head()
Out[45]:
gender past_3_years_bike_related_purchases job_title job_industry_category wealth_segment deceased_indicator owns_car tenure address postcode state country property_valuation Rank Value name age age_group
0 Male 86 General Manager Manufacturing Mass Customer N Yes 14 45 Shopko Center 4500 QLD Australia 6 1 1.718750 Chickie Brister 65 The Elderly
1 Male 69 Structural Engineer Property Mass Customer N No 16 14 Mccormick Park 2113 NSW Australia 11 1 1.718750 Morly Genery 52 Middle-Aged People
2 Female 10 Senior Cost Accountant Financial Services Affluent Customer N No 10 5 Colorado Crossing 3505 VIC Australia 5 1 1.718750 Ardelis Forrester 48 Middle-Aged People
3 Female 64 Account Representative III Manufacturing Affluent Customer N Yes 5 207 Annamark Plaza 4814 QLD Australia 1 4 1.703125 Lucine Stutt 43 Middle-Aged People
4 Female 34 Financial Analyst Financial Services Affluent Customer N No 19 115 Montana Place 2093 NSW Australia 9 4 1.703125 Melinda Hadlee 57 Middle-Aged People
In [46]:
#drop some unneccessary columns
new_df = NewCustomerList.drop(['deceased_indicator', 'age_group', 'country', 'name', 'address', 'Rank', 'Value'], axis = 1)
In [47]:
#transform categorical variables into columns with numerical values by label encoding
le = LabelEncoder()
new_df['gender'] = le.fit_transform(new_df['gender'])
new_df['job_title'] = le.fit_transform(new_df['job_title'])
new_df['job_industry_category'] = le.fit_transform(new_df['job_industry_category'])
new_df['wealth_segment'] = le.fit_transform(new_df['wealth_segment'])
new_df['owns_car'] = le.fit_transform(new_df['owns_car'])
new_df['state'] = le.fit_transform(new_df['state'])
new_df
Out[47]:
gender past_3_years_bike_related_purchases job_title job_industry_category wealth_segment owns_car tenure postcode state property_valuation age
0 1 86 71 5 2 1 14 4500 1 6 65
1 1 69 165 6 2 0 16 2113 0 11 52
2 0 10 139 2 0 0 10 3505 2 5 48
3 0 64 4 5 0 1 5 4814 1 1 43
4 0 34 68 2 0 0 19 2093 0 9 57
... ... ... ... ... ... ... ... ... ... ... ...
978 1 60 108 2 0 0 9 2200 0 7 63
979 1 22 144 3 2 0 6 2196 0 10 21
980 0 17 32 2 0 1 15 4702 1 2 68
981 1 30 67 2 2 1 19 4215 1 2 70
982 1 56 158 6 2 1 14 2010 0 9 67

983 rows × 11 columns

3.2. Fit 1000 new customers data in to best model¶

In [48]:
target_new = rf.predict(new_df)
In [49]:
target_proba = []
for x in rf.predict_proba(new_df):
     target_proba.append(x[1])
In [50]:
target_new = pd.DataFrame(target_new, columns = ['target'])
NewCustomerList['target'] = target_new['target']
NewCustomerList['target_proba'] = target_proba
NewCustomerList.sort_values(by = 'target_proba', ascending=False).head(5)
Out[50]:
gender past_3_years_bike_related_purchases job_title job_industry_category wealth_segment deceased_indicator owns_car tenure address postcode state country property_valuation Rank Value name age age_group target target_proba
207 Female 12 Social Worker Health High Net Worth N No 19 06 Old Gate Park 2144 NSW Australia 9 206 1.137500 Jenelle Fearnill 64 The Elderly 1 0.70
59 Male 47 Cost Accountant Financial Services High Net Worth N Yes 17 06936 Bobwhite Circle 2257 NSW Australia 7 57 1.375000 Lorrie Antonelli 39 Adults 1 0.70
746 Male 69 Health Coach III Health High Net Worth N No 17 7870 Stuart Crossing 2090 NSW Australia 7 755 0.640000 Meade Bampton 40 Middle-Aged People 1 0.69
4 Female 34 Financial Analyst Financial Services Affluent Customer N No 19 115 Montana Place 2093 NSW Australia 9 4 1.703125 Melinda Hadlee 57 Middle-Aged People 1 0.69
458 Male 42 Business Systems Development Analyst Financial Services Mass Customer N No 19 6218 Delladonna Parkway 4115 QLD Australia 8 466 0.890000 Thorn Stigers 50 Middle-Aged People 1 0.68
In [ ]: