Created personalized product recommendation engine predictive models and Market basket analysis for products to be purchased by existing and new customers. Created Multi output ensemble classifier model with MAP score of 0.39.
2. IMPORTANT TAKEAWAYS FROM
PREVIOUS WORK
• Data consists of more than 13 million records for 48 variables; 956,645
unique customers
• There are 24 products offered by Santander bank
• There are 3 major segments of customers – VIP, INDIVIDUAL,
UNIVERSITY GRADUATES
• Data has some seasonality component
• Majority of the customers are inactive
2
4. DATA PREPROCESSING
• Removed Columns with 99% empty records
• Converted columns into relevant data types
• One Hot Encoding of Categorical variables
• Missing value imputation for Household Income
4
13. May 2015 May 2016
Key New products : Credit Card, Debit Card, Payroll, Pension, E-Account, Current Accounts
PRODUCT PURCHASE TRENDS OF EXISTING CUSTOMERS
13
16. 16
Let’s say, we recommended 7 products and 1st, 4th, 5th, 6th product was correct. so the
result would look like — 1, 0, 0, 1, 1, 1, 0.
In this case,
1. The precision at 1 will be: 1/1 = 1
2. The precision at 2 will be: 0
3. The precision at 3 will be: 0
4. The precision at 4 will be: 2/4 = 0.5
5. The precision at 5 will be: 3/5 = 0.6
6. The precision at 6 will be: 4/6 = 0.66
7. The precision at 7 will be: 0
Average Precision will be: 1 + 0 + 0 + 0.5 + 0.6 + 0.66 + 0 /4 = 0.69
MODEL METRIC - MAP@7
19. BUSINESS ADD-ONS FROM THE
ANALYSIS
● Personalized recommendations as a service
● Reduce Manual Workload
● Transform Prospects to Clients
● Customer Satisfaction
● Increased Revenue
19
20. BUSINESS STRATEGIES
● Start with Lowest hanging fruit
● Understand the Channels customers willing to cross-sold through
● Make use of the data obtained from browsing patterns of the
customer on the website and use selective advertisement
● Offer more digital banking services and capitalize on Zero Moment of
Truth by introducing the sales team at the right time
20