Capstone Final Project

COUNTDOWN BEGINS

0 : 00 : 000

🍀Good Luck Finishing Presentation In Five Minutes.🍀

\(\color{red}{\text{ Capstone Project : Telecom Churn Analytics}}\)

\(\color{red}{\text{☎️Telecom Churn Industry ☎️}}\)

\(\color{red}{\text{2.Project Definition}}\)

Definition of Churn:

• Cancelation of Subscription.
• Closure of an active account
• Non Renewal of Contract/service Agreement
• Decision to switch carriers/provider
• End of relationship between subscriber & provider.
• Getting divorced with your loved/hate ones.

\(\color{red}{\text{3.Hypothesis}}\)

3.Hypothesis for Telecom Project.

• Supervised Classification problem
• The driving hypothesis behind this Project would be to find all the variables that play key role in minimizing customer defections.

• How can Telecom reduce the Churn Ratio so that Telecom Company stays in business.

• Predict if the customer will Churn or Not looking at past historical data. Trying to find the associated probabilities?

\(\color{red}{\text{4.Project Workflow}}\)

\(\color{red}{\text{5.The Data Overview}}\)

• Source : Kaggle
  
  
• 3333 Entries &
  
  
• 21 variables
  
  
• 1 Categorical
  
  
• 20 Numeric Variables

\(\color{red}{\text{7.Clean-up & Manipulation}}\)

PRE-PRCOESSING

• Customer Satisfaction are hardest to capture.
• didn’t have features like Product ratings & customer support calls.
• look at the dimensions
• Review the structure of Input datasets
• Find out if there is any missing values
• How features are interrelated or correlated
• Evaluate presence of outliers if any.
• Trim White Space from Voice_Mail_Plan and then recode to Zero and One.
• Trim White Space from International_Plan then recode to Zero and One.

• \(\color{green}{\text{Active Minutes ✅}}\)
  
  
• \(\color{green}{\text{Active Charges ✅}}\)
  
  
  

\(\color{red}{\text{9.Model Building Machine learning}}\)

• handle imbalanced classes
  
  
  
  
  
  
  
  
  

👍10.Choose Best Model👍

Training Sets:70%👍 Validation Sets:15% AUC Score👍👍

11.🌳Random Forest 🌳

• large difference between the training and validation scores: High variance
• increase the complexity of the model rather than having simple model by adding some new features.
• try Reducing Regularization and change model architecture to reduce underfitting.
• I look into inspire new feature ideas and try to make my model complicated to decrease variance.

12.💡Positive Feature Importance Score 💡

🎛️Hyperparameter Tunning 🎛️

• max_depth controls the depth of the tree in Random Forest.
• Change the number of variables that are sampled at each split.
• Use the AUC as the comparison metric.

⛩️13. Model Evaluation ROC Curve ⛩️

14. Conclusion:

• 1. We all know that it would have been lot easier to interpret linear classifier than non linear models.
• 2. The biggest disadvantage of being linear would be to have high bias towards fitting non linear model.
• 3. One of the advantage of Random Forest was it can work with highly corelated features. It can deal with non-linearrties much better than logistic Regression.
• 4. To make prediction it uses multiple decission trees which is more effective than single decission tree.
• 5. Then for each observation that it encounters from each node, it choose their category by taking the majority vote of different trees.
• 6.The data with low bias & high variance (non-linear relationships) can be handled properly & efficiently.

15. Conclusion:

• 7. Senstivity (Measure of True Positive Rate(TPR)/Recall (TP/(TP+FN)): 74%
• 8. That means with this model we can predcit 74% of time who will Churn before hand with current features.
• 9. Lets Look at AUC which is 0.832 which is much better than our baseline model 0.5

16. Reference

• https://www.kaggle.com/c/churn-analytics-bda/data. [Datasets]

• Ali.J(2009)“Predicting Customer Churn in Telecommunication Service provider”. Retrieved from http://www.diva-portal.org/smash/get/diva2:1020047/FULLTEXT01.pdf

• Andrew Long Course Materials Retrieved from https://northeastern.blackboard.com/webapps/portal/execute/tabs/tabAction?tab_tab_group_id=_17_1

• \(\color{green}{\text{ 17.Reference to All Image Source}}\)
  
  
• FLow Diagram : Made with Vision Software
• https://emojipedia.org/person-with-folded-hands/
• https://giphy.com/
• http://www.libpng.org/pub/png/pngcode.html
• https://icons8.com/
• https://www.w3schools.com/howto/howto_js_countdown.asp

\(\color{red}{\text{18.⌛ 🙏Thank You 🙏⌛ }}\)

\(\color{red}{\text{Name: Pankaj Shah }}\)
\(\color{red}{\text{ALY-6020: Predictive Analytics}}\)
\(\color{red}{\text{Final Capstone Project}}\)
\(\color{red}{\text{Instructor Andrew Long }}\)

COUNTDOWN ENDS

0 : 00 : 000

😞😭Please Scroll to your First Slide to see how You did Time Wise. Rendering Failed.😞😭