Customer-Churn-Prediction

A web-application made with flask that predicts whether a customer will churn or not

Link to the Web Application: https://churn-prediction.onrender.com

Built with

Exploratory Data Analysis

The data preprocessing and EDA can be found in the EDA.ipynb jupyter notebook. The data was preprocessed to remove duplicates, the dropping the Null values as the percentage of Null values was small. EDA techniques used include univariate and bivariate analysis to draw inferences about how the churn is correlated to different data features.

Inferences drawn from EDA:

Customers who use Electronic Check are the higest churners
Customers with long term contracts are more likely to churn than the customers with short term contracts
Customers with no tech support churn more
Senior citizens are less likely to churn compared to non senior citizens

Model Building

This can be found in the Churn_Model.ipynb jupyter notebook The numeric values such as TotalCharges and MonthlyCharges were used without any changes and the categorical values such as gender, paymentMethod, etc. were one hot encoded

Initially the model was trained one the original preprocessed data with Decision Tree and the following results werwe obtained:

Accuracy: 0.79
Precision: 0.74
Recall: 0.71
F1 Score: 0.72

The performance of the model was not good, and was as expected because the dataset is skewed, with less samples for No Churn. Upsampling with SMOTE-ENN and retraining the model, significantly improved the performance of the model, with the following results:

Accuracy: 0.94
Precision: 0.94
Recall: 0.94
F1 Score: 0.94

Apart from decision tree, different models were trained such as Random Forest Classifier and XGBoost. On comparing the results no significant difference was found in the performance of these models.

Web Application