Skip to content

A Comparative Study in Customer Churn Prediction through Multilayer Perceptrons and Support Vector Machines

Notifications You must be signed in to change notification settings

AnveshaM/Predicting_Customer_Churn_Neural_Network

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

7 Commits
 
 
 
 
 
 

Repository files navigation

Predicting_Customer_Churn_Neural_Network

A Comparative Study in Customer Churn Prediction through Multilayer Perceptrons and Support Vector Machines The purpose of this project is to provide a critical assessment of two algorithm models used in a supervised pattern recognition job for customer churn prediction. Feedforward Multilayer Perceptron (MLP) and Support Vector Machines are the two algorithms being compared (SVM). Hyperparameters of the base models are tuned and optimised by cross-validated grid search. The best-evaluated models are compared using Confusion Matrix and Receiver Operating Curves(ROC). For this classification task, the MLP model had better performance.

Dataset - This data set is taken from Kaggle and contains information about a bank's customers, with the target variable indicating whether the customer left the bank (closed his account) or remained a customer. It contains 10001 customer data points with 14 attributes. There are no "null" values in the dataset. The two categorical features are encoded into numerical features. All the features are scaled using the MinMaxScaler(). The target column had values 0 for Not Exited and 1 for Exited. Row Number, Customer Name and Surname are dropped from the dataset as they are not needed in the analysis. There is a class imbalance in the dataset. Out of the 10001 data points, 6356 data points are customers who have not exited while 1644 data points are of customers who have exited. About 79.63% of the data belong to 'Not Exited' customers while 20.37% of the data belong to 'Exited' customers.

Jupyter Notebook Contents -

  1. Initial Exploratory Data Analysis
  2. Feature Selection using Logistic Regression
  3. Removing Class Imbalance using SMOTE + ENN hybrid technique
  4. MultiLayer Perceptron and Hyperparameter Tuning
  5. Support Vector Machines and Hyperparameter Tuning
  6. Plotting ROC curves
  7. Voting Classifier (Additional Technique to get the best out of the two models)

Repository Contents :

  1. Churn_Modelling.csv - CSV file containing the dataset.
  2. Predicting_Customer_Churn_Neural_Network.html - HTML version of the Jupyter Notebook

About

A Comparative Study in Customer Churn Prediction through Multilayer Perceptrons and Support Vector Machines

Topics

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages