Problem Statement Understanding the consumption pattern for credit cards at an individual consumer level is important for customer relationship management. This understanding allows banks to customize for consumers and make strategic marketing plans. Thus it is imperative to study the relationship between the characteristics of the consumers and their consumption patterns. Here the dataset is of some XYZ Bank that has given a sample of their customers, along with their details like age, gender and other demographics. Also shared are information on liabilities, assets and history of transactions with the bank for each customer. In addition to the above, data has been provided for a particular set of customers' credit card spend in the previous 3 months (April, May & June) and their expected average spend in the coming 3 months (July, August & September). The average spend for different set of customers needs to be predicted in the test set for the coming 3 months. Data Dictionary id Unique ID for every Customer account_type Account Type – current or saving gender Gender of customer-M or F age Age of customer region_code Code assigned to region of residence (has order) cc_cons_apr Credit card spend in April dc_cons_apr Debit card spend in April cc_cons_may Credit card spend in May dc_cons_may Debit card spend in May cc_cons_jun Credit card spend in June dc_cons_jun Debit card spend in June cc_count_apr Number of credit card transactions in April cc_count_may Number of credit card transactions in May cc_count_jun Number of credit card transactions in June dc_count_apr Number of debit card transactions in April dc_count_may Number of debit card transactions in May dc_count_jun Number of debit card transactions in June card_lim Maximum Credit Card Limit allocated personal_loan_active Active personal loan with other bank vehicle_loan_active Active Vehicle loan with other bank personal_loan_closed Closed personal loan in last 12 months vehicle_loan_closed Closed vehicle loan in last 12 months investment_1 DEMAT investment in june investment_2 fixed deposit investment in june investment_3 Life Insurance investment in June investment_4 General Insurance Investment in June debit_amount_apr Total amount debited for April credit_amount_apr Total amount credited for April debit_count_apr Total number of times amount debited in april credit_count_apr Total number of times amount credited in april max_credit_amount_apr Maximum amount credited in April debit_amount_may Total amount debited for May credit_amount_may Total amount credited for May credit_count_may Total number of times amount credited in May debit_count_may Total number of times amount debited in May max_credit_amount_may Maximum amount credited in May debit_amount_jun Total amount debited for June credit_amount_jun Total amount credited for June credit_count_jun Total number of times amount credited in June debit_count_jun Total number of times amount debited in June max_credit_amount_jun Maximum amount credited in June loan_enq Loan enquiry in last 3 months (Y or N) emi_active Monthly EMI paid to other bank for active loans cc_cons (Target) Average Credit Card Spend in next three months
Evaluation Metric Submissions are evaluated on Root Mean Squared Logarithmic Error(RMSLE) between the predicted credit card consumption and the observed target.
Approach At first, I conducted exploratory data analysis of the dataset to gain a deeper understanding of the data. Next, I did feature engineering to create new variables.Then I tried some scikit-learn models out of which XGBoost and Random Forest gave good RMSLE. In the end I created a stacked model of those two with Linear Regression and it has been selected as the final model.
RMSLE: 115.02