• Dream Housing Finance is a company that deals with all kinds of home loan.
• They want to automate the process the loan eligibility.
• In this project we have developed various models including Logistic Regression, Tree Based Model and XG Boost.
• Best Accuracy on Real Test DataSet was achived with Logistic Regression.

• Introduction
• Problem Statement
• Importing & Installing Libraries
• Data Acquisition & Description
• Data Pre-Processing
• Exploratory Data Analysis
• Data Preparation
• Modelling
• Summary
• Contact Information

• Our Client for this Project is Dream Housing Finance. The company deals in all kinds of home loans.
• They have a presence across all urban, semi-urban and rural areas.
• Customer first applies for home loan and after that company validates the customer eligibility for loan.
• Company wants to automate the loan eligibility process (real time) based on customer detail provided while filling online application form.
• To automate this process, they have provided a dataset to identify the customers segments that are eligible for loan amount so that they can specifically target these customers.

• The Current process is a manual process and time consuming.
• Company has hired us as a Data Scientist to resolve their problem.

We have used several libraries in our project, including:

• pandas
• numpy
• matplotlib
• seaborn
• sklearn

Train Set:

• The train dataset contains 614 rows and 12 columns.
• Loan_Status is our target variable.
• The training dataset can be downloaded from here.

Test Set:

• The test dataset contains 367 rows and 11 columns.
• The test set doesnt contain the Loan_Status column.
• It needs to be predicted for the test set.
• The test dataset can be downloaded from here.

Our Dataset has the following features along with there dtypes

• We have imputed our missing values.
• Data types of Dependents and Credit History are changed.
• We have replaced '3+' with '3' in 'dependent' feature.

Observations

• Distribution of Gender is fairly unequal. Lots of Male as compared to females.
• People with marital status 'Married' are more compared to 'Unmarried'.
• More Graduate people than not Graduate.
• Most of the people are not self employed.
• Outliers present in Applicant Income,Coapplicant Income,Loan Amount and Loan Amount Term.
• Very few people with Credit History 0
• In terms of Property Area Semiurban > Urban > Rural.
• People having Loan are almost double of people not having Loan.
• Using the chi2 contigency test we saw -
  • Gender and Loan Status are Independent with p = 0.66
  • Education and Loan_status are dependent with p =0.03
  • Credit_History has 100% dependency with Loan_status with p =0.00
  • Self_employed and Loan_status are Independent with p =0.93
  • Property Area and Loan Staus are higly dependent on each other with p =0.0

• We have created a new column of Total Income = ApplicantIncome + CoapplicantIncome.
• We have converted our Categorical features into Numerical using Label Encoder.
• To bring Continous features on the same scale we have used Standard Scaler.

• We have splitted our Data into Dependent and Independent features.(x and y)
• We have used sklearn train_test_split to split our Data into x_train,x_test,y_train,y_test.

We have deployed the following models -

• We tried performing hyperparamter tuning on Ada Boost and XG Boost and saw an increase in Accuracy.
• We made an app using Gradio to show whether a person is eligible for Loan or not.

• After removing all the missing values and correcting inconsistencies we applied our model on Real Test Dataset.

• Logistic Regression was performing best with 77.77% accuracy on Analytics Vidya Hackathon.

• We have secured top 5% position on Analytics Vidya Hackathon with the above approch.

• We can further tune our other models for better accuracy.

• We can also try removing few features which are less important in predicting Loan Status and see if our model performance increases.

• We can try and work on the outliers in our continous features using domain knowledge or with client's inputs.

  Thanks.

Open for questions and suggestions.

You can contact me on Email - [email protected] LinkedIn - www.linkedin.com/in/anmol-arora-71371318