table content

Table of contents
---------------------------
Telecom Churn Case Study

Problem Statement
-----------------------
Step 1 : Data Understanding And Importing importnt libraries
Step 2 : Data Cleaning And missing Values
Handling missing values in rows
Tag churners
There is very little percentage of churn rate. We will take care of the class imbalance later.
Outliers treatment
Derive new features
Deriving new column decrease_mou_action
Deriving new column decrease_rech_num_action
Deriving new column decrease_rech_amt_action
EDA
Univariate analysis
Analysis of the minutes of usage MOU (churn and not churn) in the action phase
Bivariate analysis
Analysis of churn rate by the decreasing recharge amount and volume based cost in the action phase
Dropping few derived columns, which are not required in further analysis
Train-Test Split
Dealing with data imbalance
Scaling the test set
Model with PCA
Performing PCA with 50 components
Logistic regression with PCA
Logistic regression with optimal C
Prediction on the test set
Model summary
Support Vector Machine(SVM) with PCA
Hyperparameter tuning
Plotting the accuracy with various C and gamma values
Train set
Decision tree with PCA
Model with optimal hyperparameters
Desicison tree
Random forest with PCA
Prediction on the train set
Final conclusion with PCA
Without PCA
Feature Selection Using RFE
Model-1 with RFE selected columns
Model 2
Model-3
MODEL-4
MODEL-5
MODEL-6
Hence, we can conclued that *Model-6 log_no_pca will be the final model.
Analysis of the above curve
Confusion metrics
Plotting the ROC Curve (Trade off between sensitivity & specificity)
Testing the model on the test set
Business recomendation