Commercial banks receive a lot of applications for credit cards. Many of them get rejected for many reasons, like high loan balances, low income levels, or too many inquiries on an individual’s credit report, for example. Manually analyzing these applications is mundane, error-prone, and time-consuming (and time is money!). Luckily, this task can be automated with the power of machine learning and pretty much every commercial bank does so nowadays. In this notebook, we will build an automatic credit card approval predictor using machine learning techniques, just like the real banks do.
We’ll use the Credit Card Approval dataset from the UCI Machine Learning Repository. The structure of this notebook is as follows:
First, we will start off by loading and viewing the dataset.
We will see that the dataset has a mixture of both numerical and non-numerical features, that it contains values from different ranges, plus that it contains a number of missing entries.
We will have to preprocess the dataset to ensure the machine learning model we choose can make good predictions.
After our data is in good shape, we will do some exploratory data analysis to build our intuitions.
Finally, we will build a machine learning model that can predict if an individual’s application for a credit card will be accepted.
First, loading and viewing the dataset. We find that since this data is confidential, the contributor of the dataset has anonymized the feature names.
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15
0 b 30.83 0.000 u g w v 1.25 t t 1 f g 00202 0 +
1 a 58.67 4.460 u g q h 3.04 t t 6 f g 00043 560 +
2 a 24.50 0.500 u g q h 1.50 t f 0 f g 00280 824 +
3 b 27.83 1.540 u g w v 3.75 t t 5 t g 00100 3 +
4 b 20.17 5.625 u g w v 1.71 t f 0 f s 00120 0 +
Inspecting the applications
The output may appear a bit confusing at its first sight, but let’s try to figure out the most important features of a credit card application. The features of this dataset have been anonymized to protect the privacy, but this blog gives us a pretty good overview of the probable features. The probable features in a typical credit card application are Gender, Age, Debt, Married, BankCustomer, EducationLevel, Ethnicity, YearsEmployed, PriorDefault, Employed, CreditScore, DriversLicense, Citizen, ZipCode, Income and finally the ApprovalStatus. This gives us a pretty good starting point, and we can map these features with respect to the columns in the output.
As we can see from our first glance at the data, the dataset has a mixture of numerical and non-numerical features. This can be fixed with some preprocessing, but before we do that, let’s learn about the dataset a bit more to see if there are other dataset issues that need to be fixed.
# Print summary statisticscc_apps_description = cc_apps.describe()print(cc_apps_description)print('\n')# Print DataFrame informationcc_apps_info = cc_apps.info()print(cc_apps_info)print('\n')# Inspect missing values in the datasetprint(cc_apps.tail(17))
2 7 10 14
count 690.000000 690.000000 690.00000 690.000000
mean 4.758725 2.223406 2.40000 1017.385507
std 4.978163 3.346513 4.86294 5210.102598
min 0.000000 0.000000 0.00000 0.000000
25% 1.000000 0.165000 0.00000 0.000000
50% 2.750000 1.000000 0.00000 5.000000
75% 7.207500 2.625000 3.00000 395.500000
max 28.000000 28.500000 67.00000 100000.000000
<class 'pandas.core.frame.DataFrame'>
RangeIndex: 690 entries, 0 to 689
Data columns (total 16 columns):
# Column Non-Null Count Dtype
--- ------ -------------- -----
0 0 690 non-null object
1 1 690 non-null object
2 2 690 non-null float64
3 3 690 non-null object
4 4 690 non-null object
5 5 690 non-null object
6 6 690 non-null object
7 7 690 non-null float64
8 8 690 non-null object
9 9 690 non-null object
10 10 690 non-null int64
11 11 690 non-null object
12 12 690 non-null object
13 13 690 non-null object
14 14 690 non-null int64
15 15 690 non-null object
dtypes: float64(2), int64(2), object(12)
memory usage: 86.4+ KB
None
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15
673 ? 29.50 2.000 y p e h 2.000 f f 0 f g 00256 17 -
674 a 37.33 2.500 u g i h 0.210 f f 0 f g 00260 246 -
675 a 41.58 1.040 u g aa v 0.665 f f 0 f g 00240 237 -
676 a 30.58 10.665 u g q h 0.085 f t 12 t g 00129 3 -
677 b 19.42 7.250 u g m v 0.040 f t 1 f g 00100 1 -
678 a 17.92 10.210 u g ff ff 0.000 f f 0 f g 00000 50 -
679 a 20.08 1.250 u g c v 0.000 f f 0 f g 00000 0 -
680 b 19.50 0.290 u g k v 0.290 f f 0 f g 00280 364 -
681 b 27.83 1.000 y p d h 3.000 f f 0 f g 00176 537 -
682 b 17.08 3.290 u g i v 0.335 f f 0 t g 00140 2 -
683 b 36.42 0.750 y p d v 0.585 f f 0 f g 00240 3 -
684 b 40.58 3.290 u g m v 3.500 f f 0 t s 00400 0 -
685 b 21.08 10.085 y p e h 1.250 f f 0 f g 00260 0 -
686 a 22.67 0.750 u g c v 2.000 f t 2 t g 00200 394 -
687 a 25.25 13.500 y p ff ff 2.000 f t 1 t g 00200 1 -
688 b 17.92 0.205 u g aa v 0.040 f f 0 f g 00280 750 -
689 b 35.00 3.375 u g c h 8.290 f f 0 t g 00000 0 -
Splitting the dataset into train and test sets
Now, we will split our data into train set and test set to prepare our data for two different phases of machine learning modeling: training and testing. Ideally, no information from the test data should be used to preprocess the training data or should be used to direct the training process of a machine learning model. Hence, we first split the data and then preprocess it.
Also, features like DriversLicense and ZipCode are not as important as the other features in the dataset for predicting credit card approvals. To get a better sense, we can measure their statistical correlation to the labels of the dataset. But this is out of scope for this project. We should drop them to design our machine learning model with the best set of features. In Data Science literature, this is often referred to as feature selection.
# Import train_test_splitfrom sklearn.model_selection import train_test_split# Drop the features 11 and 13cc_apps = cc_apps.drop([11, 13], axis=1)# Split into train and test setscc_apps_train, cc_apps_test = train_test_split(cc_apps, test_size=0.33, random_state=42)
Handling the missing values (part i)
Now we’ve split our data, we can handle some of the issues we identified when inspecting the DataFrame, including:
Our dataset contains both numeric and non-numeric data (specifically data that are of float64, int64 and object types). Specifically, the features 2, 7, 10 and 14 contain numeric values (of types float64, float64, int64 and int64 respectively) and all the other features contain non-numeric values.
The dataset also contains values from several ranges. Some features have a value range of 0 - 28, some have a range of 2 - 67, and some have a range of 1017 - 100000. Apart from these, we can get useful statistical information (like mean, max, and min) about the features that have numerical values.
Finally, the dataset has missing values, which we’ll take care of in this task. The missing values in the dataset are labeled with ‘?’, which can be seen in the last cell’s output of the second task.
Now, let’s temporarily replace these missing value question marks with NaN.
# Import numpyimport numpy as np# Replace the '?'s with NaN in the train and test setscc_apps_train = cc_apps_train.replace('?', np.NaN)cc_apps_test = cc_apps_test.replace('?', np.NaN)
Handling the missing values (part ii)
We replaced all the question marks with NaNs. This is going to help us in the next missing value treatment that we are going to perform.
An important question that gets raised here is why are we giving so much importance to missing values? Can’t they be just ignored? Ignoring missing values can affect the performance of a machine learning model heavily. While ignoring the missing values our machine learning model may miss out on information about the dataset that may be useful for its training. Then, there are many models which cannot handle missing values implicitly such as Linear Discriminant Analysis (LDA).
So, to avoid this problem, we are going to impute the missing values with a strategy called mean imputation.
# Impute the missing values with mean imputation# Identify numeric columns in the DataFramenumeric_columns = cc_apps_train.select_dtypes(include=[np.number]).columns# Impute the missing values with mean imputation for numeric columns onlycc_apps_train[numeric_columns] = cc_apps_train[numeric_columns].fillna(cc_apps_train[numeric_columns].mean())cc_apps_test[numeric_columns] = cc_apps_test[numeric_columns].fillna(cc_apps_test[numeric_columns].mean())# Count the number of NaNs in the datasets and print the counts to verifyprint(cc_apps_train.isnull().sum())print(cc_apps_test.isnull().sum())
We have successfully taken care of the missing values present in the numeric columns. There are still some missing values to be imputed for columns 0, 1, 3, 4, 5, 6 and 13. All of these columns contain non-numeric data and this is why the mean imputation strategy would not work here. This needs a different treatment.
We are going to impute these missing values with the most frequent values as present in the respective columns. This is good practice when it comes to imputing missing values for categorical data in general.
# Iterate over each column of cc_apps_trainfor col in cc_apps_train.columns:# Check if the column is of object typeif cc_apps_train[col].dtypes =='object':# Impute with the most frequent value cc_apps_train = cc_apps_train.fillna(cc_apps_train[col].value_counts().index[0]) cc_apps_test = cc_apps_test.fillna(cc_apps_train[col].value_counts().index[0])# Count the number of NaNs in the dataset and print the counts to verifyprint(cc_apps_train.isnull().sum())print(cc_apps_test.isnull().sum())
There is still some minor but essential data preprocessing needed before we proceed towards building our machine learning model. We are going to divide these remaining preprocessing steps into two main tasks:
Convert the non-numeric data into numeric.
Scale the feature values to a uniform range.
First, we will be converting all the non-numeric values into numeric ones. We do this because not only it results in a faster computation but also many machine learning models (like XGBoost) (and especially the ones developed using scikit-learn) require the data to be in a strictly numeric format. We will do this by using the get_dummies() method from pandas.
# Convert the categorical features in the train and test sets independentlycc_apps_train = pd.get_dummies(cc_apps_train)cc_apps_test = pd.get_dummies(cc_apps_test)# Reindex the columns of the test set aligning with the train setcc_apps_test = cc_apps_test.reindex(columns= cc_apps_train.columns, fill_value=0)
Preprocessing the data (part ii)
Now, we are only left with one final preprocessing step of scaling before we can fit a machine learning model to the data.
Now, let’s try to understand what these scaled values mean in the real world. Let’s use CreditScore as an example. The credit score of a person is their creditworthiness based on their credit history. The higher this number, the more financially trustworthy a person is considered to be. So, a CreditScore of 1 is the highest since we’re rescaling all the values to the range of 0-1.
# Import MinMaxScalerfrom sklearn.preprocessing import MinMaxScaler# Segregate features and labels into separate variablesX_train, y_train = cc_apps_train.iloc[:, :-1].values, cc_apps_train.iloc[:, [-1]].valuesX_test, y_test = cc_apps_test.iloc[:, :-1].values, cc_apps_test.iloc[:, [-1]].values# Instantiate MinMaxScaler and use it to rescale X_train and X_testscaler = MinMaxScaler(feature_range=(0, 1))rescaledX_train = scaler.fit_transform(X_train)rescaledX_test = scaler.transform(X_test)
Fitting a logistic regression model to the train set
Essentially, predicting if a credit card application will be approved or not is a classification task. According to UCI, our dataset contains more instances that correspond to “Denied” status than instances corresponding to “Approved” status. Specifically, out of 690 instances, there are 383 (55.5%) applications that got denied and 307 (44.5%) applications that got approved.
This gives us a benchmark. A good machine learning model should be able to accurately predict the status of the applications with respect to these statistics.
Which model should we pick? A question to ask is: are the features that affect the credit card approval decision process correlated with each other? Although we can measure correlation, that is outside the scope of this notebook, so we’ll rely on our intuition that they indeed are correlated for now. Because of this correlation, we’ll take advantage of the fact that generalized linear models perform well in these cases. Let’s start our machine learning modeling with a Logistic Regression model (a generalized linear model).
# Import LogisticRegressionfrom sklearn.linear_model import LogisticRegression# Instantiate a LogisticRegression classifier with default parameter valueslogreg = LogisticRegression()# Fit logreg to the train setlogreg.fit(rescaledX_train,y_train)
LogisticRegression()
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.
LogisticRegression()
Making predictions and evaluating performance
But how well does our model perform?
We will now evaluate our model on the test set with respect to classification accuracy. But we will also take a look the model’s confusion matrix. In the case of predicting credit card applications, it is important to see if our machine learning model is equally capable of predicting approved and denied status, in line with the frequency of these labels in our original dataset. If our model is not performing well in this aspect, then it might end up approving the application that should have been approved. The confusion matrix helps us to view our model’s performance from these aspects.
# Import confusion_matrixfrom sklearn.metrics import confusion_matrix# Use logreg to predict instances from the test set and store ity_pred = logreg.predict(rescaledX_test)# Get the accuracy score of logreg model and print itprint("Accuracy of logistic regression classifier: ", logreg.score(rescaledX_test,y_test))# Print the confusion matrix of the logreg modelconfusion_matrix(y_test,y_pred)
Accuracy of logistic regression classifier: 1.0
array([[103, 0],
[ 0, 125]])
Grid searching and making the model perform better
Our model was pretty good! In fact it was able to yield an accuracy score of 100%.
For the confusion matrix, the first element of the of the first row of the confusion matrix denotes the true negatives meaning the number of negative instances (denied applications) predicted by the model correctly. And the last element of the second row of the confusion matrix denotes the true positives meaning the number of positive instances (approved applications) predicted by the model correctly.
But if we hadn’t got a perfect score what’s to be done?. We can perform a grid search of the model parameters to improve the model’s ability to predict credit card approvals.
# Import GridSearchCVfrom sklearn.model_selection import GridSearchCV# Define the grid of values for tol and max_itertol = [0.01, 0.001 ,0.0001]max_iter = [100, 150, 200]# Create a dictionary where tol and max_iter are keys and the lists of their values are the corresponding valuesparam_grid =dict(tol=tol, max_iter=max_iter)
Finding the best performing model
We have defined the grid of hyperparameter values and converted them into a single dictionary format which GridSearchCV() expects as one of its parameters. Now, we will begin the grid search to see which values perform best.
We will instantiate GridSearchCV() with our earlier logreg model with all the data we have. We will also instruct GridSearchCV() to perform a cross-validation of five folds.
We’ll end the notebook by storing the best-achieved score and the respective best parameters.
While building this credit card predictor, we tackled some of the most widely-known preprocessing steps such as scaling, label encoding, and missing value imputation. We finished with some machine learning to predict if a person’s application for a credit card would get approved or not given some information about that person.
import warningswarnings.filterwarnings("ignore") # Instantiate GridSearchCV with the required parametersgrid_model = GridSearchCV(estimator=logreg, param_grid=param_grid, cv=5)# Fit grid_model to the datagrid_model_result = grid_model.fit(rescaledX_train, y_train)# Summarize resultsbest_score, best_params = grid_model_result.best_score_, grid_model_result.best_params_print("Best: %f using %s"% (best_score, best_params))# Extract the best model and evaluate it on the test setbest_model = grid_model_result.best_estimator_print("Accuracy of logistic regression classifier: ", best_model.score(rescaledX_test,y_test))
Best: 1.000000 using {'max_iter': 100, 'tol': 0.001}
Accuracy of logistic regression classifier: 1.0
