Ecosyste.ms: Awesome
An open API service indexing awesome lists of open source software.
https://github.com/iguptashubham/ott-churn-eda-ml
Understanding why customers discontinue their subscriptions will be crucial in optimizing the user experience, reducing churn, and maximizing customer lifetime value. By using Machine learning model to predict the Customer Churn.
https://github.com/iguptashubham/ott-churn-eda-ml
data-analysis data-analysis-project data-science data-science-portfolio data-science-projects data-visualization machine-learning python
Last synced: 1 day ago
JSON representation
Understanding why customers discontinue their subscriptions will be crucial in optimizing the user experience, reducing churn, and maximizing customer lifetime value. By using Machine learning model to predict the Customer Churn.
- Host: GitHub
- URL: https://github.com/iguptashubham/ott-churn-eda-ml
- Owner: iguptashubham
- Created: 2024-07-18T07:30:07.000Z (4 months ago)
- Default Branch: main
- Last Pushed: 2024-07-18T16:55:23.000Z (4 months ago)
- Last Synced: 2024-07-19T12:01:08.528Z (4 months ago)
- Topics: data-analysis, data-analysis-project, data-science, data-science-portfolio, data-science-projects, data-visualization, machine-learning, python
- Language: Jupyter Notebook
- Homepage:
- Size: 969 KB
- Stars: 0
- Watchers: 2
- Forks: 0
- Open Issues: 0
-
Metadata Files:
- Readme: README.md
Awesome Lists containing this project
README
# Streaming OTT Subscription
For this analysis, we'll be examining customer churn data from a leading subscription-based streaming service. This industry giant boasts a vast library of movies, TV shows, and original content. Understanding why customers discontinue their subscriptions will be crucial in optimizing the user experience, reducing churn, and maximizing customer lifetime value.
# flowchart of Project
# Exploration
## Statistics
### Numerical Features
AccountAge
MonthlyCharges
TotalCharges
ViewingHoursPerWeek
AverageViewingDuration
ContentDownloadsPerMonth
UserRating
SupportTicketsPerMonth
WatchlistSize
Churn
count
243787.000000
243787.000000
243787.000000
243787.000000
243787.000000
243787.000000
243787.000000
243787.000000
243787.000000
243787.000000
mean
60.083758
12.490687
750.741017
20.502179
92.264061
24.503513
3.002713
4.504186
12.018508
0.181232
std
34.285143
4.327613
523.073273
11.243753
50.505243
14.421174
1.155259
2.872548
7.193034
0.385211
min
1.000000
4.990000
4.991154
1.000065
5.000547
0.000000
1.000007
0.000000
0.000000
0.000000
25%
30.000000
8.740000
329.147027
10.763953
48.382395
12.000000
2.000853
2.000000
6.000000
0.000000
50%
60.000000
12.500000
649.878487
20.523116
92.249992
24.000000
3.002261
4.000000
12.000000
0.000000
75%
90.000000
16.240000
1089.317362
30.219396
135.908048
37.000000
4.002157
7.000000
18.000000
0.000000
max
119.000000
19.990000
2378.723844
39.999723
179.999275
49.000000
4.999989
9.000000
24.000000
1.000000
### Categorical Features
SubscriptionType
PaymentMethod
PaperlessBilling
ContentType
MultiDeviceAccess
DeviceRegistered
GenrePreference
Gender
ParentalControl
SubtitlesEnabled
count
243787
243787
243787
243787
243787
243787
243787
243787
243787
243787
unique
3
4
2
3
2
4
5
2
2
2
top
Standard
Electronic check
No
Both
No
Computer
Comedy
Female
Yes
Yes
freq
81920
61313
121980
81737
122035
61147
49060
121930
122085
122180
## Analysis
### Churn Rate
- Not Churned Customer - 81.88%
- Churned Customer - 18.12%
### Churn by Gender
- **Churn and Gender**-
Among our users, there are more females than males. However, the churn rate for male users is slightly higher than the churn rate for female users.
### Account Age
1. **User Age and Churn:**
- Older users are more prevalent than new users.
- Churn tends to be higher among accounts with lower age (newer users).
- As account age increases, the churn rate decreases.
2. **Account Age Statistics:**
- Median account age: 60 months
- Highest account age: 120 months
3. **Churn Rates:**
- Churn rate for new users (low account age): ~32%
- Churn rate decreases to ~7% as account age increases.
### Total Monthly Charges
1. **Churn and Monthly Charges:**
- Churn rates tend to increase as monthly charges increase. Users with higher monthly charges are more likely to churn.
- Low churn occurs when monthly charges are low.
2. **Subscription Distribution:**
- There's a relatively high number of low monthly subscriptions, gradually decreasing toward higher subscription tiers.
3. **Monthly Charge Statistics:**
- Minimum monthly price: 4.99
- Maximum monthly price: 19.99
### Viewing Hours Per Week
1. **Churn and Viewing Hours:**
- Users with higher weekly viewing hours are less likely to churn compared to those with lower viewing hours.
- High churn occurs among users who view less than or equal to four hours.
- Users with viewing hours between 30 and 40 exhibit relatively stable churn.
2. **Viewing Hours Statistics:**
- Average weekly view hours: 20.5
- Minimum weekly view hours: 1
- Maximum weekly view hours: 40
### Average Viewing Duration
1. **Churn and Average View Duration:**
- Users which have the low average view duration are more likely to churn than the users which have high average duration.
- High churn occurs when the average view duration upto 40.
- there is gradually decrease in churn rate from 50
- Users with viewing hours between 125 and 175 exhibit relatively stable churn.
2. **Average view Duration Statistics :**
- Minimum is 5.
- Maximum is 179.99
- Average is 92.37.
### Content Downloads Per Month
1. **Churn and Content Downloads Per Month:**
- Users with high Content Downloads Per Month are more likely to churn compared to users with low Content Downloads Per Month.
- The sudden drop of up to 50% in some bars indicates that people download new series and shows when they are released.
- Users with high download activity likely subscribe to download new content.
- High churn occurs when the download count exceeds 10.
- There is a decrease in churn rate as the download count decreases.
- Users with download counts between 30 and 50 exhibit relatively stable churn behavior.
- Some users do not download shows or movies, possibly due to device limitations or a preference for watching on TV.
2. **Average View Duration Statistics:**
- Minimum view duration is 0.
- Maximum view duration is 49.
- The average view duration across users is 24.48.
### User Rating
1. **Churn and User Rating:**
- There is no relationship between rating and churn because rating 1 to 5 there is relatively stable churn.
- OTT streaming get relatively equal ratings.
2. **User Rating Statistics :**
- Minimum is 1.
- Maximum is 4.99
- Average is 3.002.
### Support Ticket Per month
1. **Churn and Support Ticket:**
- From the illustration, it shows that the higher the ticket generate, the users most likely to churn
- there is significantly low churn rate when there is no ticket generated. but when the ticket generated it also increaes the churn rate.
- Most churn rate are when the ticket generated 6 to 8.
2. **Support Ticket Statistics :**
- Minimum is 0.
- Maximum is 9.99
- Average is 4.5.002.
### Watch List Size
1. **Churn and Watch List Size:**
- There is no relationship between Watchlist and churn because there is relatively stable churn.
2. **User Rating Statistics :**
- Minimum is 0.
- Maximum is 24.99
- Average is 12.02.
### Churn by Subscription
1. **Churn Rate by Subscription Type**:
- The churn rate is highest among Basic subscribers, followed by Standard and Premium.
- Specifically, users who subscribe to the Basic plan tend to have a higher churn rate.
2. **Premium Subscribers**:
- Premium subscribers exhibit the lowest churn rate across all subscription types.
- This suggests that Premium subscribers are more likely to stay with the service.
3. **User Preferences**:
- Most users prefer the Premium subscription, followed by Standard and then Basic.
### Payment Method and Paperless Billing
1. **Churn and Payment Method:**
- from this illustration, Users with electronic check have the higher churn rate followed by mail check and bank transfer and credit.
- Paperless billing affecting the churn rate is relatively very low.
3. **Overall Insights:**
- users with Payment method which takes time and effort have thr highest churn rate as compared to the payment method that require less effort.
1. **Churn and Paperless Billing**:
- The diagonal represents the correlation of each variable with itself (which is always 1.0).
- The off-diagonal values represent the correlation between `Churn` and `PaperlessBilling`.
- When `PaperlessBilling` is `No`, the correlation with `Churn` is approximately -0.98, indicating a strong negative relationship. Customers who are not on paperless billing tend to have lower churn rates.
- Conversely, when `PaperlessBilling` is `Yes`, the correlation with `Churn` is around -0.91. This suggests that customers on paperless billing have a slightly higher churn rate.
2. **Overall Insights**:
- Paperless billing appears to be an important factor affecting churn behavior.
- Customers who prefer paperless billing are more likely to churn.
- As a data analyst, you might want to explore this relationship further and consider it in your churn prediction models.
### Content Type with Churn
- **Churn and Content type**:
- Movie content have the lowest churn rate followed by Tv shows and both.
- User who have subscribe both have the high churn rate.
### Churn by Genre
- **Churn and Genre**:
- Comedy have the high churn rate followed by the SCI-Fi, drama, fantasy and action have the lowest churn rate
- Most users preferred the comedy genere followed by fantasy, drama,action and SCi-fi.
### Churn by Parental Control and Subtitle Enables
- High churn rate in No parental control and no subtitle enabled.
- There is high number of users which have parent control and subtitle enabled.
### Correlation with Churn
1. Churn has a **negative correlation** with the following features:
- `AccountAge`
- `Viewinghoursperweek`
- `Averageviewduration`
- `Contentdownloadpermonth`
This implies that as these features increase, churn tends to **decrease**.
2. On the other hand, churn has a **positive correlation** with the following features:
- `Supportticketpermonth`
- `userrating`
- `watchlistsize`
This suggests that as these features increase, churn tends to **increase**.
## Conclusion
1. **Paperless Billing**:
- Customers who are not on paperless billing tend to have lower churn rates (correlation of approximately -0.98).
- However, customers on paperless billing still exhibit a slightly higher churn rate (correlation of around -0.91).
- **Insight**: Paperless billing significantly impacts churn behavior, and customers who prefer paperless billing are more likely to churn.
2. **Content Type**:
- Movie content has the lowest churn rate, followed by TV shows and both.
- Users who subscribe to both movie and TV show content have a higher churn rate.
- **Insight**: Consider tailoring content recommendations based on user preferences to reduce churn.
3. **Genre Preferences**:
- Comedy content has the highest churn rate, while genres like Sci-Fi, drama, fantasy, and action have lower churn rates.
- Most users prefer comedy, followed by fantasy, drama, action, and Sci-Fi.
- **Insight**: Personalizing content offerings based on genre preferences can improve retention.
4. **Parental Control and Subtitles**:
- High churn occurs among users with no parental control and no subtitle enabled.
- Many users have both parental control and subtitles enabled.
- **Insight**: Enhancing parental control features and ensuring subtitle availability may positively impact retention.
5. **Correlations**:
- Negative correlations:
- AccountAge, Viewinghoursperweek, Averageviewduration, and Contentdownloadpermonth are associated with lower churn rates.
- Positive correlations:
- Supportticketpermonth, userrating, and watchlistsize are associated with higher churn rates.
- **Insight**: These correlations can guide targeted strategies to reduce churn (e.g., improving customer support or enhancing content recommendations).
# Pipeline of Churn Predicting Machine learning Model
