2-Day National Level AI Hackathon At NIE,Mysuru.



What is Delinquency telecom model ?

Delinquency is a condition that arises when an activity or situation does not occur at its scheduled (or expected) date i.e., it occurs later than expected.

In this project we have data of 190001 users of  a telecom company, the data includes information such as.

1. for how may years user is using the network.

2. how many times he takes loan in 3 months.

3. average amount he pays back in 3 months.

4. and will he be able to pay the loan in time or not.

This is a classification problem.

Libraries used in this project are

  • Pandas.
  • Sci-kit Learn.

Data split techniques used is K-FOLD with 10 folds

  • Accuracy model used is confusion metrics.

Feature Selection And Extraction

Age on cellular network

Customers registered in network for less than 5 years.

Customer registered in network for more then 5 years.

Daily amount spent from main account, averaged over last 90 days

less than 0.

In between 0 to 3645.

In between greater than 3645.

Number of days till last recharge of main account

  • Less than 0.
  • In between 0 to 7 days.
  • Greater than 7 days.

Amount of recharge

  • In between 0 to 770.
  • Greater than 770.

Number of times main account recharge, in 90 days

In between 0 to 8.

Greater than 8.

Number of loans taken by user in last 90 days

In between 0 to 223.

Greater than 223.

Maximum amount of loan taken by user in last 90 days

In between 0 to 6.

Greater than 12.

Average payback time in days over last 90 days

In between 0 to 10.

Greater than 10.



The data is split into 10 subsets and accuracy is evaluated.

1. logistic regression.

Accuracy 87 %.

2. Decision trees.

Accuracy  90%

SVM Using 20% of Training Data

Model training time 10 minutes.

For 50% of data it take more than two hours.

Accuracy of model 88% for 20% of test data.

Team Name: DIVINE AI

Srujan Kachhwaha

Vaibhav Mishra

Rohan Baghel

Aman Gupta

Davine ai
Share This Post To Your Friends

Similar Posts

Leave a Reply

Your email address will not be published. Required fields are marked *