A leading Retail Bank wanted to know the target customer profile, so, that the marketing campaignu00a0 would be effective. The target is to disburse personal load to existing and new customers.
Various key data were gathered. Using Ai u2013 ML, developed a model that would predict the probability(likelihood) that a customer would take a loan, given a set of customer profile data.
- Data collected during campaign can be used to Analyze and improve the efficiency.
- Key variables are
- Age, Job, A/C balance, Previous Loan, Marital status
- Analyses the data and clean the raw data
- Compare key variables with one another to establish any correlation
- Establish key variables that impact the outcome using feature engineering
- Build a ML model using Python using key variables
- Train using the key variables
- Test it against the Known outcome
- Use multiple models(ensemble) to arrive at optimum model
- Use the model to predict customers for targeted campaign
- Total records > 50k
- Variables [ ‘serial_number’, ‘age_in_years’, ‘job_description’, ‘marital_status’, ‘education_details’, ‘has_default’, ‘balance_in_account’, ‘housing_status’, ‘previous_loan’, ‘phone_type’, ‘date’, ‘month_of_year’, ‘call_duration’, ‘campaign_contacts’, ‘days_passed’, ‘previous_contact’, ‘poutcome_of_campaign’, ‘outcomeu2019]
- Feature engineered Variables – ‘age_in_years’, ‘job_description’, ‘education_details’, ‘has_default’, ‘balance_in_account’, ‘housing_status’, ‘previous_loan’, ‘call_duration’, ‘campaign_contacts’, ‘days_passed’, ‘previous_contact’, ‘poutcome_of_campaign’, ‘outcomeu2019
Check more on https://public.tableau.com/profile/sivakumar.d4364#!/vizhome/Campaign-Analytics-Bank/Dashboard-1
- From the data one can see Job has an influence on the outcome,
- So is the balance, having u2018Zerou2019 balance not necessarily will result in u2018Yesu2019 for a loan
- One cannot conclude from the data which influencers the most
- Using Machine learning it is possible to predict with reasonable accuracy
Model u2013 Logistic Regression
- Model built has mean absolute error 0.030 (0.0 is the best)
- This can be improved by better data variables and large data
- ML also enables targeted campaign and also increases efficiency
- Employeeu2019s performance can be measured more reliably
A model has been validated and finalized, a set of coefficients were chosen to predict the outcome given the chosen variables.
This can be deployed as a service or even in Excel sheet, that could do the math to get the outcome