Udacity Data Science nano degree Capstone project
Predict the user churn for some Song streaming company
The dataset seems to belong to some digital music streaming Platforms. Whenever a user interacts with the music streaming service the logs get generated like timestamp and the activity page(i.e logOut, like, logIn etc). There are basically types of user 1. free -tier and 2. subscription-based. so based on the activity of the user we need to figure out which user is likely to churn out. So, some kind of discounts can be rolled out to those specific users.
In this project, we will try to find out the likeliness that a user will continue to use the music streaming App based on past user behavior activity. So that the risk of losing potential users can be minimized.
This project uses the following software and Python libraries:
- Python3
- Spark
- Pyspark
- pandas
- Matplotlib
- Seaborn
Sparkify.ipnyb: This files contains all the code related to this project.
.PNG: These files contains the plots for Data Exploration.
final_dataset.CSV: This file contains the cleaned data with features and target Column as label.
Gradient Boosted Tree has the best performance with the F1 score of 85 approx.
Since the dataset was highly unbalanced we tried evaluating the model with under sampled dataset. but there wasn't enough improvemrnt in the model performance.
gradient boosted Tree Evaluation Metrics
| metric | value |
|---|---|
| Precision | 0.8333 |
| Recall | 0.8695 |
| F1 | 0.8510 |
I have written a blog Explaining the techniques and approch for solving this Challenge blog
NOTE: the dateset mini_sparkify_event_data.json was large enough, so I was unable to upload it to the repo due to Github size limits .