Machine Learning can Mitigate Risk for Lenders

LendingClub is an American online credit marketplace that connects borrowers and investors. It offers credit solutions such as personal loans, business loans, and refinancing solutions to borrowers. The investors are banks and institutional investors who can filter and purchase loans through specific credit attributes that fit their investment and risk preferences. With more than 3 million members, LendingClub facilitates billions in loans each year.

The loan origination process starts when a borrower completes a loan application online. LendingClub performs an initial screening based on grade using the borrower’s credit and income data. LendingClub then services the loan once investors make a lending decision.

While this business model saves investors money on marketing and loan origination cost, the return on investment is negatively impacted if the borrower defaults because most loans are unsecured – they have no collateral backing.

Business Understanding

An institutional investor/lender would like to purchase high-quality unsecured loans from LendingClub. Their goal is to create the right loan investment strategies and mix that minimizes risk and maximizes the overall returns from this loan portfolio. They want to predict the loan portfolio is profitable while pre-purchasing it.

“So answer the question – who shall we buy?”

An important consideration is that LendingClub does not require collateral for unsecured loans. Therefore, should a borrower default, the possibility of recovering the outstanding principal and interest is very low and the overall return of the portfolio is adversely impacted if such an account is eventually designated as charged-off. Therefore, reliably predicting if a loan will be fully-paid or charged-off would reduce credit risk and the associated financial loss for the lender.

  • A loan is charged-off if the company believes it is unlikely to be collected as the borrower has substantially defaulted on payments. After a loan is charged off, the lender could sell the debt to a third-party collections agency that would attempt to collect on the delinquent account. A borrower owes the debt until it is paid off, settled, discharged in a bankruptcy proceeding, or in case of legal proceedings, becomes too old due to the statute of limitations.
  • A loan is deemed fully-paid if the lender has collected all loan payments, including both the principal balance and accrued interest.

Credit risk is the possibility of financial loss that may arise when a borrower fails to make the required payment on loans to a lender. Financial loss could include the loss of principal and interest, disruption in cash flo, and increased collection costs.

Lenders gauge the creditworthiness of potential borrows using the five C’s of credit.

  1. Character, also called the credit history, measures the borrowers’ reputation and history of repaying debt. Factors include length of employment, credit score, liens, and judgment reports.
  2. Capacity measures the ability of the borrower to repay the loan by comparing income against debts. Lower debt-to-income ratio is usually preferred.
  3. Conditions consider the interest rate, loan term, and loan amount.
  4. Capital measures the contribution that the borrower puts towards a potential investment. Examples are down payments for mortgages and auto financing. LendingClub does not evaluate loan applications based on this criterion.
  5. Collateral are assets that borrowers use to secure a loan. They give the assurance that the lender can recover all or a portion of the loan in the event of default. Collateral is not required for most of LendingClub loan products.

LendingClub collects data such as employment length, annual income, FICO score, number of open accounts, public record bankruptcies, tax liens, inquiries in the last 6 months, number of satisfactory accounts, and other loan specific information from all loan applications. 

Data Description

The raw data was 584,862 rows, 152 columns, and 23,116,937 missing values:

Column nameDescription
acc_now_delinqThe number of accounts on which the borrower is now delinquent.
acc_open_past_24mthsNumber of trades opened in the past 24 months.
addr_stateThe state provided by the borrower in the loan application
all_utilBalance to credit limit on all trades
annual_incThe self-reported annual income provided by the borrower during registration.
annual_inc_jointThe combined self-reported annual income provided by the co-borrowers during registration
application_typeIndicates whether the loan is an individual application or a joint application with two co-borrowers
avg_cur_balAverage current balance of all accounts
bc_open_to_buyTotal open to buy on revolving bank cards.
bc_utilRatio of total current balance to high credit/credit limit for all bankcard accounts.
chargeoff_within_12_mthsNumber of charge-offs within 12 months
collection_recovery_feePost charge off collection fee
collections_12_mths_ex_medNumber of collections in 12 months excluding medical collections
delinq_2yrsThe number of 30+ days past-due incidences of delinquency in the borrower’s credit file for the past 2 years
delinq_amntThe past-due amount owed for the accounts on which the borrower is now delinquent.
descLoan description provided by the borrower
dtiA ratio calculated using the borrower’s total monthly debt payments on the total debt obligations, excluding mortgage and the requested LC loan, divided by the borrower’s self-reported monthly income.
dti_jointA ratio calculated using the co-borrowers’ total monthly payments on the total debt obligations, excluding mortgages and the requested LC loan, divided by the co-borrowers’ combined self-reported monthly income
earliest_cr_lineThe month the borrower’s earliest reported credit line was opened
emp_lengthEmployment length in years. Possible values are between 0 and 10 where 0 means less than one year and 10 means ten or more years. 
emp_titleThe job title supplied by the Borrower when applying for the loan. The employer title replaces employer name for all loans listed after 9/23/2013
fico_range_highThe upper boundary range the borrower’s FICO at loan origination belongs to.
fico_range_lowThe lower boundary range the borrower’s FICO at loan origination belongs to.
funded_amntThe total amount committed to that loan at that point in time
funded_amnt_invThe total amount committed by investors for that loan at that point in time
gradeLC assigned loan grade
home_ownershipThe home ownership status provided by the borrower during registration or obtained from the credit report. The values are RENT, OWN, MORTGAGE, and OTHER.
idA unique LC assigned ID for the loan listing.
il_utilRatio of total current balance to high credit/credit limit on all install acct
initial_list_statusThe initial listing status of the loan. Possible values are – W, F
inq_fiNumber of personal finance inquiries
inq_last_12mNumber of credit inquiries in past 12 months
inq_last_6mthsThe number of inquiries in past 6 months (excluding auto and mortgage inquiries)
installmentThe monthly payment owed by the borrower if the loan originates
int_rateInterest Rate on the loan
issue_dThe month which the loan was funded
last_credit_pull_dThe most recent month LC pulled credit for this loan
last_fico_range_highThe upper boundary range the borrower’s last FICO pulled belongs to.
last_fico_range_lowThe lower boundary range the borrower’s last FICO pulled belongs to.
last_pymnt_amntLast total payment amount received
last_pymnt_dLast month payment was received
loan_amntThe listed amount of the loan applied for by the borrower. If at some point in time, the credit department reduces the loan amount, then it will be reflected in this value.
loan_statusCurrent status of the loan, the target feature.
max_bal_bcMaximum current balance owed on all revolving accounts
member_idA unique LC assigned Id for the borrower member.
mo_sin_old_il_acctMonths since oldest bank instalment account opened
mo_sin_old_rev_tl_opMonths since oldest revolving account opened
mo_sin_rcnt_rev_tl_opMonths since most recent revolving account opened
mo_sin_rcnt_tlMonths since most recent account opened
mort_accNumber of mortgage accounts
mths_since_last_delinqThe number of months since the borrower’s last delinquency
mths_since_last_major_derogMonths since most recent 90-day or worse rating
mths_since_last_recordThe number of months since the last public record
mths_since_rcnt_ilMonths since most recent instalment accounts opened
mths_since_recent_bcMonths since the most recent bank card account opened
mths_since_recent_bc_dlqMonths since most recent bank card delinquency
mths_since_recent_inqMonths since the most recent inquiry
mths_since_recent_revol_delinqMonths since the most recent revolving delinquency
next_pymnt_dNext scheduled payment date
num_accts_ever_120_pdNumber of accounts ever 120 or more days past due
num_actv_bc_tlNumber of currently active bankcard accounts
num_actv_rev_tlNumber of currently active revolving trades
num_bc_satsNumber of satisfactory bankcard accounts
num_bc_tlNumber of bankcard accounts
num_il_tlNumber of installment accounts
num_op_rev_tlNumber of open revolving accounts
num_rev_acctsNumber of revolving accounts
num_rev_tl_bal_gt_0Number of revolving trades with balance >0
num_satsNumber of satisfactory accounts
num_tl_120dpd_2mNumber of accounts currently 120 days past due (updated in past 2 months)
num_tl_30dpdNumber of accounts currently 30 days past due (updated in past 2 months)
num_tl_90g_dpd_24mNumber of accounts 90 or more days past due in last 24 months
num_tl_op_past_12mNumber of accounts opened in past 12 months
open_accThe number of open credit lines in the borrower’s credit file.
open_acc_6mNumber of open trades in last 6 months
open_il_12mNumber of instalment accounts opened in past 12 months
open_il_24mNumber of instalment accounts opened in past 24 months
open_act_ilNumber of currently active instalment trades
open_rv_12mNumber of revolving trades opened in past 12 months
open_rv_24mNumber of revolving trades opened in past 24 months
out_prncpRemaining outstanding principal for total amount funded
out_prncp_invRemaining outstanding principal for portion of total amount funded by investors
pct_tl_nvr_dlqPercent of trades never delinquent
percent_bc_gt_75Percentage of all bank card accounts > 75% of limit.
policy_codePublicly available policy_code=1
New products not publicly available policy_code=2
pub_recNumber of derogatory public records
pub_rec_bankruptciesNumber of public record bankruptcies
purposeA category provided by the borrower for the loan request
pymnt_planIndicates if a payment plan is put in place for the loan
recoveriesPost charge off gross recovery
revol_balTotal credit revolving balance
revol_utilRevolving line utilization rate, or the amount of credit the borrower is using relative to all available revolving credit
sub_gradeLC assigned loan subgrade
tax_liensNumber of tax liens
termThe number of payments on the loan. Values are in months and can be either 36 or 60.
titleThe loan title provided by the borrower
tot_coll_amtTotal collection amounts ever owed
tot_cur_balTotal current balance of all accounts
tot_hi_cred_limTotal high credit/credit limit
total_accThe total number of credit lines currently in the borrower’s credit file
total_bal_ex_mortTotal credit balance excluding mortgage
total_bal_ilTotal current balance of all instalment accounts
total_bc_limitTotal bankcard high credit/credit limit
total_cu_tlNumber of finance trades
total_il_high_credit_limitTotal instalment high credit/credit limit
total_pymntPayments received to date for total amount funded
total_pymnt_invPayments received to date for portion of total amount funded by investors
total_rec_intInterest received to date
total_rec_late_feeLate fees received to date
total_rec_prncpPrincipal received to date
total_rev_hi_lim  Total revolving high credit/credit limit
urlURL for the LC page with listing data.
verification_statusIndicates if income was verified by LC, not verified, or if the income source was verified
verified_status_jointIndicates if the co-borrowers’ joint income was verified by LC, not verified, or if the income source was verified
zip_codeThe first 3 numbers of the zip code provided by the borrower in the loan application.
revol_bal_jointSum of revolving credit balance of the co-borrowers, net of duplicate balances
sec_app_fico_range_lowFICO range (high) for the secondary applicant
sec_app_fico_range_highFICO range (low) for the secondary applicant
sec_app_earliest_cr_lineEarliest credit line at time of application for the secondary applicant
sec_app_inq_last_6mthsCredit inquiries in the last 6 months at time of application for the secondary applicant
sec_app_mort_acc Number of mortgage accounts at time of application for the secondary applicant
sec_app_open_acc Number of open trades at time of application for the secondary applicant
sec_app_revol_utilRatio of total current balance to high credit/credit limit for all revolving accounts
sec_app_open_act_ilNumber of currently active instalment trades at time of application for the secondary applicant
sec_app_num_rev_accts Number of revolving accounts at time of application for the secondary applicant
sec_app_chargeoff_within_12_mthsNumber of charge-offs within last 12 months at time of application for the secondary applicant
sec_app_collections_12_mths_ex_medNumber of collections within last 12 months excluding medical collections at time of application for the secondary applicant
sec_app_mths_since_last_major_derogMonths since most recent 90-day or worse rating at time of application for the secondary applicant
hardship_flagFlags if the borrower is on a hardship plan
hardship_typeDescribes the hardship plan offering
hardship_reasonDescribes the reason the hardship plan was offered
hardship_statusDescribes if the hardship plan is active, pending, canceled, completed, or broken
deferral_termAmount of months that the borrower is expected to pay less than the contractual monthly payment amount due to a hardship plan
hardship_amountThe interest payment that the borrower has committed to make each month while they are on a hardship plan
hardship_start_dateThe start date of the hardship plan period
hardship_end_dateThe end date of the hardship plan period
payment_plan_start_dateThe day the first hardship plan payment is due. For example, if a borrower has a hardship plan period of 3 months, the start date is the start of the three-month period in which the borrower is allowed to make interest-only payments.
hardship_lengthThe number of months the borrower will make smaller payments than normally obligated due to a hardship plan
hardship_dpdAccount days past due as of the hardship plan start date
hardship_loan_statusLoan Status as of the hardship plan start date
orig_projected_additional_accrued_interestOriginal projected additional interest amount that will accrue for the given hardship payment plan as of the Hardship Start Date. It will be null if the borrower has broken their hardship payment plan.
hardship_payoff_balance_amountThe payoff balance amount as of the hardship plan start date
hardship_last_payment_amountThe last payment amount as of the hardship plan start date
disbursement_methodThe method by which the borrower receives their loan. Possible values are: CASH, DIRECT_PAY
debt_settlement_flagFlags whether or not the borrower, who has charged-off, is working with a debt-settlement company.
debt_settlement_flag_dateThe most recent date that the Debt_Settlement_Flag has been set  
settlement_statusThe status of the borrower’s settlement plan. Possible values are: COMPLETE, ACTIVE, BROKEN, CANCELLED, DENIED, DRAFT
settlement_dateThe date that the borrower agrees to the settlement plan
settlement_amountThe loan amount that the borrower has agreed to settle for
settlement_percentageThe settlement amount as a percentage of the payoff balance amount on the loan
settlement_termThe number of months that the borrower will be on the settlement plan

To understand the data better, features such as ‘employment length, annual income, FICO score, number of open accounts, public record bankruptcies, tax liens, inquiries in the last 6 months, and number of satisfactory accounts’ were plotted against the feature ‘Loan Status’ to provide a visual representation of the relationship that exists between the dependent (target) and independent variables.

It was observed that borrowers with 10+ years length of employment are most likely to fully pay their loans. Other employment lengths also exhibit a relationship. Other features showed a similar relationship to ‘Loan Status’.

Another analytic was to plot the ‘Purpose’ of the loan against the ‘Loan Status’.

It shows that debt consolidation loans are most likely to be paid off. A similar relationship was observed for all purposes.

The ‘loan status in percentage’ diagram shows that 15.70% of the loans were charged off while 84.30% were fully paid. This was a good observation as we will have to balance this for the purposes of machine learning later.

A model that can reliably predict whether a loan will be fully paid or charged-off could help reduce the financial loss to the investor’s credit portfolio. The ‘Loan_Status’ is the target variable. Other features, also called dependent variables, will be used to predict the Loan_Status.

A ML model was created to predict the likelihood that a borrower who passes the initial screening but may default on the loan subsequently. The dependent variable (also called target feature) was ‘Loan_Status’. It had two categories, fully-paid or charged-off. 

Braintoy mlOS was used to wrangle the data and create the ML models. In conjugation, Tableau was used for data visualization.

Data Engineering

Data Wrangling

This is the first step in machine learning that removes errors, inconsistencies, incomplete, and irrelevant information. 

First, columns with less relevant data were deleted. Examples are data that cannot be estimated/obtained before the loan is disbursed, columns with the same values, duplicate data, or columns that show a similar relationship to Loan_Status as other columns. 

List of Deleted Features








































































































Second, all rows with the value of ‘Joint’ on the ‘Application type’ column and all columns with information about joint applications were deleted because borrowers who struggle to qualify for loans on their own may usually add a co-borrower to qualify and obtain favorable terms. Including these records may lead to a low-quality model. Therefore, it was important to exclude joint applications and rather have a separate analysis for that portion of the dataset. 

Third, substrings “%” and “months” were deleted from interest rate and term respectively and the data type converted to numeric. At this stage, there were 39 columns and 452,069 samples. 

After these data wrangling operations, it was observed that the ‘loan status’ column (the dependent/target variable) had 90,817 records for charged-off and 361,252 records for fully-paid loans.

Because the number of fully-paid loans far outweigh the charged-off loans, training a machine learning model on this dataset will lead to a biased model. 

There are several techniques to resolve an imbalanced data problem such as creating synthetic data for the minority class, but because the sample size was large, random resampling (undersampling) to delete 97% of rows where the value equals ‘Fully paid’ and 90% of rows where the value equals “Charged Off” was applied on the loan status column.

This reduced the sample size to 18,294 (9,212 Fully Paid and 9,082 Charged Off).

This now became a balanced dataset that is fit for machine learning.

Feature Selection

This is the process of selecting the input features and the target feature and finding the relative importance of such features to predict the ‘Loan Status’. 

Feature importance is a technique that is used to identify crucial features for an efficient model. Features with higher scores will have a higher effect on the model than features with a lower score. In addition to the benefit listed above, feature selection saves time and money by eliminating irrelevant features with lower scores. 

The table below shows the score for the top ten features. In addition to these top ten features, the employment length, annual income, loan purpose, and initial list status were also selected to train the ML model.

S/NInput featureImportance score

Feature Preprocessing

This is the process of converting raw data into a specific format to make it work well for machine learning. 

Categorical to numerical refactoring were applied on columns with categorical values such as sub_grade, emp_length, verification_status, loan_status, purpose, initial_list_status, and application_type. 

Cross Validation

This is the process of splitting a dataset into the training and validation datasets. The training dataset is used to train the ML models and the validation dataset is used to evaluate the model’s performance. 

For this analysis, 80% of the dataset was used to train the model while 20% was used for validation.

Model Building

The purpose of the model is to predict if a loan will be fully paid or charged off. This is a binary decision – yes or no. Hence, it is a classification problem.

Classification is a supervised approach in machine learning where the algorithms learn from the provided data input and can then classify new data from what they learnt. 

Automated machine learning was performed that created several models using various classification algorithms. 

Of all the algorithms, the Extra Tree Classifier provided the highest accuracy of 95.41%. It was selected for publishing.

There were several other challenger models but the top three are Random Forest Classifier with accuracies of 95.35%, Quadratic Discriminant Analysis with an accuracy of 94.34%, and XGB classifier with an accuracy of 94.21%.

Model Evaluation

Confusion Matrix

Confusion matrix is a table that visualizes the performance of a model by showing the number of False Positives, False Negatives, True Positives, and True Negatives. A good matrix has larger numbers on the diagonal with a darker shade of blue and small numbers on the lighter shade of blue. 

In the diagram below, charged-off loans are labeled as 0 and fully-paid loans are labeled as 1. 1,800 (49.1%) loans were correctly predicted as charged off and 1,700 (46.3%) loans were correctly predicted as paid-off. Only 79 (2.2%) were incorrectly predicted as charged-off, and only 89 (2.4%) were incorrectly predicted as fully paid. Therefore, the percentage of loans that will eventually be charged off can be reduced from 15.7% (from the ‘Loan status in percentage’ diagram) to 2.4% by what the model learnt.

ROC Curve

The Receiver Operating Characteristic (ROC) curve is used to show how well a model can distinguish between True Values and Predicted Values. This is an indication of if the model has truly learned or simply memorized from the training data.

The more the curve fits the top left corner of the plot, the better the classification process.

This diagram shows that the Area Under the Curve (AUC) is 0.99, and it closely hugs the top left corner of the plot.

This is a good model!

Model Governance

A reviewer has to now double-check the published model/(s) for biases, inefficiencies, and inaccuracies. Model/(s) may be accepted as-is or may be rejected with a request to the modeler to improve it and publish a better version for review.

This maker-checker method brings prudence to machine learning operations (MLOps). Automated documentation makes it easy.

Jaspreet Gill of Braintoy was the coach for this internship. She reviewed the published model.

To start with, the Training and the Test sample were checked for ‘skewness’.

Note that the performance metrics were automatically calculated by comparing predicted results against the test data provided to the algorithms. But what if the data to train the model vs. the data to test it were dissimilar! The performance metrics will have lesser meaning then. The ‘skew’ gives a visual hint that there might be something wrong.

The performance scores could be relied on if the distribution of data is similar for each feature in the training and test set.

It was observed that the distribution of training vs. test for each feature were similar. In addition, the target variable was balanced. The performance metrics could be relied on.

This was a supervised classification problem that shows an accuracy of 95.41%. This seemed to be a good model. But while accuracy is the simplest measure to evaluate model performance, the other evaluation criteria such as ROC curve, Confusion Matrix, F1-score, Hamming Loss, Precision, Recall, and the Jaccard Score are also important.

  • Precision answers the question – what proportion of positive identifications were actually correct? Recall answers the question – What proportion of actual positives were correctly classified? Both Precision and Recall scores were 0.95 indicating that the model is performing well.
  • The F1 score is the harmonic mean of Precision and Recall, a summary indicator. A score of 95.41 indicated good performance.
  • The Hamming Loss indicates the fraction of records that were incorrectly predicted. While for binary classification problems such as this, it is just 1 minus accuracy, but it becomes more relevant for multi-class classifications as hamming loss averages the ‘loss’ from each class across the dataset. A score of 0.05 shows that the published model has less ‘loss’.
  • The Jaccard Score is a measure of similarity between predictions and the test set and their intersection and union. Measured between 0 and 1, higher score indicates that predictions overlap more with the test set. In this case, a score of 0.91 indicated good performance.

Based on these performance metrics as well as the ROC and the Confusion Matrix described in the previous section, the model was approved for production use. 

The modeler can now deploy it as an API/microservice.

Model Deployment

After the model is reviewed and approved, the next step is to integrate it into a production environment for practical business decision-making. 

The first step is to create an app wherein the model will be containerized. 

The app was created and named LendingClubLoanApp.

The second step is to select the newly created app under Deployable Apps. 

In this step, the LendingClubLoanApp and the version v.2-v.934 of the model were selected, the version that was approved in the Model Governance step.

The third step is to containerize the model in the app.

There are two options – the app can be downloaded as a .zip file to a local computer or deployed to the cloud as an API/Microservice.

Option 1: Download app to local computer

Select “Download App” on the model deployment page.

Select HTML/Javascript as the code type, input the file name, and confirm the download.

The file was saved as “LendingClubLoanApp” on the local computer.

It could now be opened on any installed browser.

For security, the API keys and access token are required to run the app. Note that an ‘unauthorized’ error message is displayed in the model response log if the correct API key and access token are not entered. 

To generate the API key and access token, the LendingClubLoanApp and “Manage Apps” buttons on the Model Deployment page were selected.

Then on the App Manager page, the ‘API Keys’ and ‘+ Add API KEY’ were selected.

Client ID and Client Secret were created on the API Key and Access Token page. 

The public key and secret key were generated and saved as key pairs.

On the App Manager page, the saved Key Pair must be shared with the user for validation. It is only activated when shared.

The Key Pair was shared with myself to test the AI application.

Clicking on the View icon could show the API keys and Access Token.

The keys were copied to the appropriate sections on the HTML/Javascript page previously downloaded to the local computer.

The model is now ready for use.

To test the app, some known values were put to check that the model predicts that the loan will be paid.

  • loan_amnt: 5000
  • int_rate: 20.0
  • sub_grade: D2
  • emp_length: 1 year
  • annual_inc: 42000.0
  • verification_status: Not Verified
  • purpose: other
  • dti: 6.46
  • inq_last_6mths: 0
  • mths_since_last_delinq: 21
  • open_acc: 4
  • initial_list_status: w
  • last_fico_range_high: 669
  • last_fico_range_low: 665

As a second test, alternate values were put to check that the model predicts that the loan will be charged-off.

  • loan_amnt: 5000
  • int_rate: 20.0
  • sub_grade: D2
  • emp_length: 1 year
  • annual_inc: 42000.0
  • verification_status: Not Verified
  • purpose: other
  • dti: 46.46
  • inq_last_6mths: 0
  • mths_since_last_delinq: 21
  • open_acc: 4
  • initial_list_status: w
  • last_fico_range_high: 469
  • last_fico_range_low: 445

Option 2: Deploy the app on the cloud

In this option, the app will run on the mlOS cluster and can be made accessible to anyone in the world with whom the API Keys are shared.

To deploy the model to the mlOS cluster, ‘Deploy’ was selected under model deployment and mlOS-mlCluster-01 as the destination type.

Once deployed, the status of the LendingClubLoanApp changed to “Running” which tells that the app is ready for use.


Select the app from the My Apps page to open it.

Realtime Prediction

It accepts one set of inputs at a time and predicts an output. 

When one set of test values were input, the output was predicted that the loan would be fully paid.

For another set of test inputs, the app predicted an output that the loan will be charged off.

Data Scoring

Rather than typing the inputs one by one, data scoring allows records to be uploaded in bulk for scoring and prediction.

In data scoring, the probability of each output class is calculated and ranked, and the output class with the highest probability is selected. 

The file for the validation set was selected and scored.

As seen in the first row in the above screenshot, the probability of the loan being fully paid is 0.03 while the probability of the loan being charged off is 0.97. Hence, the loan is predicted to be charged off. 

All rows in the dataset were scored and appropriate predictions were observed.


Predicting loans that will likely be charged off reduces financial losses to the portfolio mix of an investor. It was observed that the ML models could reduce the exposure to risky loans from 15.4% to 2.4%.

“Get 13% more from your investment by buying loans that predict Fully Paid and avoiding loans that predict Charged Off.”

The business environment is dynamic. Situations change. If the variables used to build the ML model change, the effectiveness of the model also reduces causing model decay. It is prudent to periodically review changes to the data characteristics and keep the model current with improved versions.


Tolu Alade is a finance professional with almost a decade of experience in compliance, commercial, and retail banking. 

She has an MBA in Finance, a Bachelor’s degree in Economics, and is a certified full stack developer with proficiency in Python. She has expertise in data analytics, machine learning, and data science techniques.