"Building a machine learning model that Predicts a loan defaulter"
•Download as PPTX, PDF•
1 like•231 views
The bank was experiencing high levels of non-performing assets (NPAs) due to many loan defaulters. A machine learning model was developed to predict loan applicants' default status with 90% accuracy. The model was trained on data of over 500,000 customers with 45 variables, including payment history, credit utilization, collections, and application details. When applied to new applicants, the model correctly predicted 68% of defaulters. Important variables for identifying defaulters included last payment amount, credit limits, interest paid, balances, collections, loan grade, and applicant's state of residence.
"Building a machine learning model that Predicts a loan defaulter"
- 1. atom D sciences Building Machine learning applications for Enterprises Industry : Finance/Micro-Finance/Loans Case study : Predicting which applicant will be a Loan defaulter
- 2. Problem Statement The Bank Indessa has not done well in last 3 quarters. Their NPAs (Non Performing Assets) have reached all time high. After careful analysis, it was found that the majority of NPA was contributed by loan defaulters. We developed a machine learning Model that predicted the applicants default status with almost 90% accuracy.
- 3. Data set Number of Rows(Customers) : 532428 Number of Columns (variables) : 45 TARGET : 0 means Default ;1 means- No default
- 4. Train –Test Division (85% -15%)
- 5. Sl.no variable class levels missing value percent zeroes percent 1 acc_now_delinq integer 0 0.003052 99.533740 2 addr_state factor 51 0 0.000000 3 annual_inc numeric 425943 0.000704 0.000000 4 application_type factor 2 0 0.000000 5 batch_enrolled factor 104 19.920506 0.000000 6 collection_recovery_fee numeric 425943 0 97.360914 7 collections_12_mths_ex_med integer 0 0.018547 98.664375 8 delinq_2yrs integer 0 0.003052 80.722773 9 desc factor 70639 0 0.000000 10 dti numeric 425943 0 0.047189 11 emp_length factor 11 5.045276 0.000000 12 emp_title factor 190125 0.00047 0.000000 13 funded_amnt integer 0 0 0.000000 14 funded_amnt_inv numeric 425943 0 0.026999 15 grade factor 7 0 0.000000 16 home_ownership factor 6 0 0.000000 17 initial_list_status factor 2 0 0.000000 18 inq_last_6mths integer 0 0.003052 56.156340 19 int_rate numeric 425943 0 0.000000 20 last_week_pay factor 98 0 0.000000 21 loan_amnt integer 0 0 0.000000 22 mths_since_last_delinq integer 0 51.138533 0.189697 23 mths_since_last_major_derog integer 0 75.024827 0.016904 24 mths_since_last_record integer 0 84.580566 0.137108 25 open_acc integer 0 0.003052 0.000939 26 pub_rec integer 0 0.003052 84.713917 27 purpose factor 14 0 0.000000 28 pymnt_plan factor 2 0 0.000000 29 recoveries numeric 425943 0 97.230615 30 revol_bal numeric 425943 0 0.373524 31 revol_util numeric 425943 0.054467 0.394419 32 sub_grade factor 35 0 0.000000 33 target integer 0 0 76.343783 34 term factor 2 0 0.000000 35 title factor 39694 0 0.000000 36 tot_coll_amt integer 0 7.898944 79.062926 37 tot_cur_bal numeric 425943 7.898944 0.014086
- 6. Applied Machine learning algorithms and Build Predictive Model on 4.32 lakh customers
- 8. Predicting 68% defaulters correctly 90% overall accuracy
- 9. Which variables are important to find the defaulters?
- 10. Dictionary Last_week_pay,total_revolving credit limit, Interest received till date, total current balace of all accounts, Total collection amount ever owed, Sub_grade,Addr_state
- 11. Contact : V Raviteja Valluri, Founder & Data Scientist, Atom D Sciences & Analytics Pvt Ltd, raviteja@atomdsciences.com, +91-8501903007 www.atomdsciences.com