Python - IA and Predictions

Case: Customer Credit Score

Project Description and KPIs

You were hired by a bank to be able to define the credit score of customers. You need to analyze all the bank's customers and, based on this analysis, create a model that can read the customer's information and automatically say his credit score: Bad, Ok, Good

Libraries: Pandas, Numpy, Scikit Learn

Data 📊 - Description and modeling

Step 1 - Importing the data source

import pandas as pd
table = pd.read_csv("clientes.csv")

Step 2 - Visualization of the data base

display(table)

	id_cliente	mes	idade	profissao	salario_anual	num_contas	num_cartoes	juros_emprestimo	num_emprestimos	dias_atraso	...	idade_historico_credito	investimento_mensal	comportamento_pagamento	saldo_final_mes	score_credito	emprestimo_carro	emprestimo_casa	emprestimo_pessoal	emprestimo_credito	emprestimo_estudantil
0	3392	1	23.0	cientista	19114.12	3.0	4.0	3.0	4.0	3.0	...	265.0	21.465380	alto_gasto_pagamento_baixos	312.494089	Good	1	1	1	1	0
1	3392	2	23.0	cientista	19114.12	3.0	4.0	3.0	4.0	3.0	...	266.0	21.465380	baixo_gasto_pagamento_alto	284.629162	Good	1	1	1	1	0
2	3392	3	23.0	cientista	19114.12	3.0	4.0	3.0	4.0	3.0	...	267.0	21.465380	baixo_gasto_pagamento_medio	331.209863	Good	1	1	1	1	0
3	3392	4	23.0	cientista	19114.12	3.0	4.0	3.0	4.0	5.0	...	268.0	21.465380	baixo_gasto_pagamento_baixo	223.451310	Good	1	1	1	1	0
4	3392	5	23.0	cientista	19114.12	3.0	4.0	3.0	4.0	6.0	...	269.0	21.465380	alto_gasto_pagamento_medio	341.489231	Good	1	1	1	1	0
...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...
99995	37932	4	25.0	mecanico	39628.99	4.0	6.0	7.0	2.0	23.0	...	378.0	24.028477	alto_gasto_pagamento_alto	479.866228	Poor	1	0	0	0	1
99996	37932	5	25.0	mecanico	39628.99	4.0	6.0	7.0	2.0	18.0	...	379.0	24.028477	alto_gasto_pagamento_medio	496.651610	Poor	1	0	0	0	1
99997	37932	6	25.0	mecanico	39628.99	4.0	6.0	7.0	2.0	27.0	...	380.0	24.028477	alto_gasto_pagamento_alto	516.809083	Poor	1	0	0	0	1
99998	37932	7	25.0	mecanico	39628.99	4.0	6.0	7.0	2.0	20.0	...	381.0	24.028477	baixo_gasto_pagamento_alto	319.164979	Standard	1	0	0	0	1
99999	37932	8	25.0	mecanico	39628.99	4.0	6.0	7.0	2.0	18.0	...	382.0	24.028477	alto_gasto_pagamento_medio	393.673696	Poor	1	0	0	0	1

100000 rows × 25 columns

Step 3 - Data Cleaning | error handling

Preparing the data source to the IA

print(table.info())

<class 'pandas.core.frame.DataFrame'>
RangeIndex: 100000 entries, 0 to 99999
Data columns (total 25 columns):
 #   Column                    Non-Null Count   Dtype  
---  ------                    --------------   -----  
 0   id_cliente                100000 non-null  int64  
 1   mes                       100000 non-null  int64  
 2   idade                     100000 non-null  float64
 3   profissao                 100000 non-null  object 
 4   salario_anual             100000 non-null  float64
 5   num_contas                100000 non-null  float64
 6   num_cartoes               100000 non-null  float64
 7   juros_emprestimo          100000 non-null  float64
 8   num_emprestimos           100000 non-null  float64
 9   dias_atraso               100000 non-null  float64
 10  num_pagamentos_atrasados  100000 non-null  float64
 11  num_verificacoes_credito  100000 non-null  float64
 12  mix_credito               100000 non-null  object 
 13  divida_total              100000 non-null  float64
 14  taxa_uso_credito          100000 non-null  float64
 15  idade_historico_credito   100000 non-null  float64
 16  investimento_mensal       100000 non-null  float64
 17  comportamento_pagamento   100000 non-null  object 
 18  saldo_final_mes           100000 non-null  float64
 19  score_credito             100000 non-null  object 
 20  emprestimo_carro          100000 non-null  int64  
 21  emprestimo_casa           100000 non-null  int64  
 22  emprestimo_pessoal        100000 non-null  int64  
 23  emprestimo_credito        100000 non-null  int64  
 24  emprestimo_estudantil     100000 non-null  int64  
dtypes: float64(14), int64(7), object(4)
memory usage: 19.1+ MB
None

# a. Label Encoder - transform text in number
from sklearn.preprocessing import LabelEncoder
from sklearn.model_selection import train_test_split

encoder = LabelEncoder()
table["profissao"] = encoder.fit_transform(table["profissao"])
table["mix_credito"] = encoder.fit_transform(table["mix_credito"])
table["comportamento_pagamento"] = encoder.fit_transform(table["comportamento_pagamento"])
#display(table)

# b. Separating data in 2: training and testing data
x = table.drop(["score_credito", "id_cliente"], axis=1)
y = table["score_credito"]
x_training, x_test, y_training, y_test = train_test_split(x,y, test_size=0.3, random_state=1)

Step 4 - Create IA models to predict the credit score

# 2 models: Decision Tree and KNN

# a. Import
from sklearn.ensemble import RandomForestClassifier
from sklearn.neighbors import KNeighborsClassifier

# b. Create
model_decision_tree = RandomForestClassifier()
model_knn = KNeighborsClassifier()

# c. Train
model_decision_tree.fit(x_training, y_training)
model_knn.fit(x_training, y_training)

KNeighborsClassifier()

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.

Step 5 - Calculate the accuracy of the models | Compare them and choose the best

from sklearn.metrics import accuracy_score
prediction_decision_tree = model_decision_tree.predict(x_test)
prediction_knn = model_knn.predict(x_test.to_numpy())

print(accuracy_score(y_test, prediction_decision_tree))
print(accuracy_score(y_test, prediction_knn))

/opt/homebrew/lib/python3.11/site-packages/sklearn/base.py:464: UserWarning: X does not have valid feature names, but KNeighborsClassifier was fitted with feature names
  warnings.warn(

0.8251666666666667
0.7324

83% vs. 73% - The best model is the decision tree

Step 6 - Predict the credit score of new customers

We'll use decision tree model to predictions

table_new_customers = pd.read_csv("novos_clientes.csv")
table_new_customers["profissao"] = encoder.fit_transform(table_new_customers["profissao"])
table_new_customers["mix_credito"] = encoder.fit_transform(table_new_customers["mix_credito"])
table_new_customers["comportamento_pagamento"] = encoder.fit_transform(table_new_customers["comportamento_pagamento"])
print(table_new_customers.info())
display(table_new_customers)

new_prediction = model_decision_tree.predict(table_new_customers)
print(new_prediction)

<class 'pandas.core.frame.DataFrame'>
RangeIndex: 3 entries, 0 to 2
Data columns (total 23 columns):
 #   Column                    Non-Null Count  Dtype  
---  ------                    --------------  -----  
 0   mes                       3 non-null      int64  
 1   idade                     3 non-null      float64
 2   profissao                 3 non-null      int64  
 3   salario_anual             3 non-null      float64
 4   num_contas                3 non-null      float64
 5   num_cartoes               3 non-null      float64
 6   juros_emprestimo          3 non-null      float64
 7   num_emprestimos           3 non-null      float64
 8   dias_atraso               3 non-null      float64
 9   num_pagamentos_atrasados  3 non-null      float64
 10  num_verificacoes_credito  3 non-null      float64
 11  mix_credito               3 non-null      int64  
 12  divida_total              3 non-null      float64
 13  taxa_uso_credito          3 non-null      float64
 14  idade_historico_credito   3 non-null      float64
 15  investimento_mensal       3 non-null      float64
 16  comportamento_pagamento   3 non-null      int64  
 17  saldo_final_mes           3 non-null      float64
 18  emprestimo_carro          3 non-null      int64  
 19  emprestimo_casa           3 non-null      int64  
 20  emprestimo_pessoal        3 non-null      int64  
 21  emprestimo_credito        3 non-null      int64  
 22  emprestimo_estudantil     3 non-null      int64  
dtypes: float64(14), int64(9)
memory usage: 684.0 bytes
None

	mes	idade	profissao	salario_anual	num_contas	num_cartoes	juros_emprestimo	num_emprestimos	dias_atraso	num_pagamentos_atrasados	...	taxa_uso_credito	idade_historico_credito	investimento_mensal	comportamento_pagamento	saldo_final_mes	emprestimo_carro	emprestimo_casa	emprestimo_pessoal	emprestimo_credito	emprestimo_estudantil
0	1	31.0	1	19300.340	6.0	7.0	17.0	5.0	52.0	19.0	...	29.934186	218.0	44.50951	1	312.487689	1	1	0	0	0
1	4	32.0	0	12600.445	5.0	5.0	10.0	3.0	25.0	18.0	...	28.819407	12.0	0.00000	2	300.994163	0	0	0	0	1
2	2	48.0	1	20787.690	8.0	6.0	14.0	7.0	24.0	14.0	...	34.235853	215.0	0.00000	0	345.081577	0	1	0	1	0

3 rows × 23 columns

['Poor' 'Good' 'Standard']

Step 7 - Main characteristics to define the score

columns = list(x_test.columns)
importance = pd.DataFrame(index=columns, data=model_decision_tree.feature_importances_)
importance = importance*100
print(importance)

                                  0
mes                        3.935890
idade                      4.204221
profissao                  3.292680
salario_anual              5.113708
num_contas                 3.655706
num_cartoes                4.556644
juros_emprestimo           8.370750
num_emprestimos            3.367472
dias_atraso                5.974806
num_pagamentos_atrasados   4.569275
num_verificacoes_credito   4.724786
mix_credito                8.581199
divida_total              10.875573
taxa_uso_credito           5.075785
idade_historico_credito    7.497293
investimento_mensal        4.854108
comportamento_pagamento    2.330494
saldo_final_mes            5.463939
emprestimo_carro           0.717342
emprestimo_casa            0.729737
emprestimo_pessoal         0.697791
emprestimo_credito         0.712744
emprestimo_estudantil      0.698057

divida_total, mix_credito and juros_emprestimo are the most important characteristics to define a poor, good or standard score

display(table)
display(table_new_customers)

	id_cliente	mes	idade	profissao	salario_anual	num_contas	num_cartoes	juros_emprestimo	num_emprestimos	dias_atraso	...	idade_historico_credito	investimento_mensal	comportamento_pagamento	saldo_final_mes	score_credito	emprestimo_carro	emprestimo_casa	emprestimo_pessoal	emprestimo_credito	emprestimo_estudantil
0	3392	1	23.0	2	19114.12	3.0	4.0	3.0	4.0	3.0	...	265.0	21.465380	1	312.494089	Good	1	1	1	1	0
1	3392	2	23.0	2	19114.12	3.0	4.0	3.0	4.0	3.0	...	266.0	21.465380	3	284.629162	Good	1	1	1	1	0
2	3392	3	23.0	2	19114.12	3.0	4.0	3.0	4.0	3.0	...	267.0	21.465380	5	331.209863	Good	1	1	1	1	0
3	3392	4	23.0	2	19114.12	3.0	4.0	3.0	4.0	5.0	...	268.0	21.465380	4	223.451310	Good	1	1	1	1	0
4	3392	5	23.0	2	19114.12	3.0	4.0	3.0	4.0	6.0	...	269.0	21.465380	2	341.489231	Good	1	1	1	1	0
...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...
99995	37932	4	25.0	11	39628.99	4.0	6.0	7.0	2.0	23.0	...	378.0	24.028477	0	479.866228	Poor	1	0	0	0	1
99996	37932	5	25.0	11	39628.99	4.0	6.0	7.0	2.0	18.0	...	379.0	24.028477	2	496.651610	Poor	1	0	0	0	1
99997	37932	6	25.0	11	39628.99	4.0	6.0	7.0	2.0	27.0	...	380.0	24.028477	0	516.809083	Poor	1	0	0	0	1
99998	37932	7	25.0	11	39628.99	4.0	6.0	7.0	2.0	20.0	...	381.0	24.028477	3	319.164979	Standard	1	0	0	0	1
99999	37932	8	25.0	11	39628.99	4.0	6.0	7.0	2.0	18.0	...	382.0	24.028477	2	393.673696	Poor	1	0	0	0	1

100000 rows × 25 columns

	mes	idade	profissao	salario_anual	num_contas	num_cartoes	juros_emprestimo	num_emprestimos	dias_atraso	num_pagamentos_atrasados	...	taxa_uso_credito	idade_historico_credito	investimento_mensal	comportamento_pagamento	saldo_final_mes	emprestimo_carro	emprestimo_casa	emprestimo_pessoal	emprestimo_credito	emprestimo_estudantil
0	1	31.0	1	19300.340	6.0	7.0	17.0	5.0	52.0	19.0	...	29.934186	218.0	44.50951	1	312.487689	1	1	0	0	0
1	4	32.0	0	12600.445	5.0	5.0	10.0	3.0	25.0	18.0	...	28.819407	12.0	0.00000	2	300.994163	0	0	0	0	1
2	2	48.0	1	20787.690	8.0	6.0	14.0	7.0	24.0	14.0	...	34.235853	215.0	0.00000	0	345.081577	0	1	0	1	0

3 rows × 23 columns

Conclusion & Insights 🎉

This project underscores the pivotal role of each data stage, evident in the successful training of classification models achieving 82% accuracy in customer scoring. It emphasizes the universal importance of data analysis in diverse industries. The insights gained extend beyond the project, offering a roadmap for insightful decision-making that drives revenue. Embracing data analysis equips individuals to uncover hidden patterns, delivering exceptional business outcomes and fueling profit enhancement.