deloitte_presents_machine_learning_challenge_predict_loan_defaulters

Competition hosted on MACHINEHACK

About

Build a machine learning model that predicts the loan defaulter.

Initially tried the ensemble models(catboost,xgboost,lightgbm).All the ensemble model's logloss values didn't show any difference. So i have tried neural network model and it gives smaller logloss value than the ensemble models and better leaderboard rank.

Competition Public LB Rank: 146/465 & Private LB Rank: 146/465

Final Score 0.3782

Evaluation Metric is Logloss.

File information

• deloitte_ml_challenge_predict_loan_defaulters.ipynb

  Packages Used,

   * Sklearn
   * catboost
   * xgboost
   * lgbm
   * keras
   * Pandas
   * klib
   * Numpy
   * Matplotlib
   * Optuna
   * shap
  

  Basic Exploratory Data Analysis

  Created Catboost classifier model and tune the hyperparameters with the optuna framework.

  Created XGboost classifier model and tune the hyperparameters with the optuna framework.

  Created Lightgbm classifier model and tune the hyperparameters with the optuna framework.

  Model interpretation with shap

  Created keras neural net model

  Model Comparison

Feature Importance Catboost

Feature Importance XGboost

Feature Importance Lightgbm

Model's Validation Logloss comparison