"Features" are values derived from the user’s data, which can be used directly in a loan decisioning flow. They are returned in the decision endpoint. Definitions of the features generated by Pngme are:
| feature | definition | response value |
|---|---|---|
| average_end_of_day_depository_balance_{t0_days}_{t1_days} | Average of daily total balance held in all depository accounts, over the previous t0 to t1 days, where the time windows are 0-30, 31-90, or 0-90 days history prior to the prediction date. | {float, null} |
| average_end_of_day_loan_balance_{t0_days}_{t1_days} | The time-average of the user’s balances, for all balances observed in SMS from known lenders, across all known lenders. The average is calculated over a period of t0 to t1 days history prior to the prediction date, where the time windows are 0-30, 31-90, or 0-90 days. | {float, null} |
| count_airtime_purchase_events_{t0_days}_{t1_days} | The number of SMS received indicating that the user purchased airtime with any telco. The counts are summed over a period of t0 to t1 days history prior to the prediction date, where the time windows are 0-30, 31-90, or 0-90 days. | {int, null} |
| count_betting_and_lottery_events_{t0_days}_{t1_days} | The number of SMS received indicating that the user engaged in betting or lottery activity. The counts are summed over a period of t0 to t1 days history prior to the prediction date, where the time windows are 0-30, 31-90, or 0-90 days. | {int, null} |
| count_insufficient_funds_events_{t0_days}_{t1_days} | The number of SMS received indicating that the user had insufficient funds to conduct a transaction. The counts are summed over a period of t0 to t1 days history prior to the prediction date, where the time windows are 0-30, 31-90, or 0-90 days. | {int, null} |
| count_loan_declined_events_{t0_days}_{t1_days} | The number of SMS received indicating that the user was declined for a loan, from any known lender. The count is summed over a period of t0 to t1 days history prior to the prediction date, where the time windows are 0-30, 31-90, or 0-90 days. | {int, null} |
| count_loan_defaulted_events_{t0_days}_{t1_days} | The number of SMS received indicating that the user defaulted on a loan, from any known lender. The count is summed over a period of t0 to t1 days history prior to the prediction date, where the time windows are 0-30, 31-90, or 0-90 days. | int |
| count_loan_missed_payment_events_{t0_days}_{t1_days} | The number of SMS received indicating that the user missed a loan payment, for a loan from any known lender. The count is summed over a period of t0 to t1 days history prior to the prediction date, where the time windows are 0-30, 31-90, or 0-90 days. | {int, null} |
| count_loan_opened_events_{t0_days}_{t1_days} | The number of SMS received indicating that the user either had a loan approved and/or had a loan disbursed, from any known lender. The counts are summed over a period of t0 to t1 days history prior to the prediction date, where the time windows are 0-30, 31-90, or 0-90 days. | {int, null} |
| count_loan_repayment_events_{t0_days}_{t1_days} | The number of SMS received indicating that the user made a loan repayment to one of their loans, from any known lender. The count is summed over a period of t0 to t1 days history prior to the prediction date, where the time windows are 0-30, 31-90, or 0-90 days. | {int, null} |
| count_of_institutions_{t0_days}_{t1_days} | The number of distinct financial institutions represented in the SMSs received by the user. The count is summed over a period of t0 to t1 days history prior to the prediction date, where the time windows are 0-30, 31-90, or 0-90 days. | {int, null} |
| count_overdraft_events_{t0_days}_{t1_days} | The number of SMS received indicating that the user had overdraft activity on an account. The count is summed over a period of t0 to t1 days history prior to the prediction date, where the time windows are 0-30, 31-90, or 0-90 days. | {int, null} |
| count_transactions_depository_{t0_days}_{t1_days} | The number of SMS received indicating that the user conducted a depository transaction. The count is summed over a period of t0 to t1 days history prior to the prediction date, where the time windows are 0-30, 31-90, or 0-90 days. | {int, null} |
| count_user_shared_device_ids_0_10 | The number of distinct users using the same device updated within the time range of 10 days history prior to the prediction date. | {int, null} |
| count_user_shared_phone_numbers_0_10 | The number of distinct users with the same phone number updated within the time range of 10 days history prior to the prediction date. | {int, null} |
| daily_average_of_stacked_loan_alerts_0_90 | The daily average number of loan-related SMS from known lenders received by a user. Loan-related SMS include, LoanDefaulted, LoanMissedPayment, LoanRepaid, LoanApproved, LoanDisbursed, LoanRepayment, and LoanRepaymentReminder, over a period of t0 to t1 days history prior to the prediction date, where the time windows are 0-30, 31-90, or 0-90 days. | {float, null} |
| daily_average_of_stacked_loan_fes_0_90 | The daily average number of lenders with whom a user has activity over a period of zero to 90 days history prior to the prediction date. Having activity means receiving one or more SMS on a given day, from a given lender, showing an active loan (the SMS stating the outstanding loan balance and/or a disbursement/repayment event). The daily average number of lenders means the mean average of lenders on a per day basis, for all days in which activity was observed with one or more lender, over the last 0 to 90 days history prior to the prediction date. | {float, null} |
| data_density_{t0_days}_{t1_days} | Ratio of active user days to total number of days in range, over the previous t0 to t1 days, where the time windows are 0-30, 31-90, or 0-90 days history prior to the prediction date. | {float, null} |
| data_recency_minutes | The time in minutes between utc_endtime and the most recent financial event or alert, as an indicator of data recency, or freshness. | {int, null} |
| difference_count_of_loans_opened_to_loans_delinquent_{t0_days}_{t1_days} | The difference between a) the number of SMS received indicating that the user was approved for a loan, or had a loan disbursed (of any loan to any lender) and b) the number of SMS received indicating that the user is past due or defaulted on a loan (of any loan to any lender). The counts for a) and b) are summed over a period of t0 to t1 days history prior to the prediction date, where the time windows are 0-30, 31-90, or 0-90 days. | {int, null} |
| difference_count_of_loans_opened_to_loans_repaid_{t0_days}_{t1_days} | Returns the count of loans approved or disbursed - loans repaid, over the previous t0 to t1 days, where the time windows are 0-30, 31-90, or 0-90 days history prior to the prediction date. The number of SMS received indicating that the user had a loan approved or disbursed, minus the number indicating they repaid a loan. The counts are summed over a period of t0 to t1 days history prior to the prediction date, where the time windows are 0-30, 31-90, or 0-90 days. | {int, null} |
| loan_sharks_alert_institutions_ratio_0_90 | Ratio of loan shark related financial events over total events over a given period, over the last 0 to 90 days history prior to the prediction date. | {float, null} |
| loan_sharks_fes_ratio_0_90 | Ratio of loan shark institutions over total seen institutions over the last 0 to 90 days history prior to the prediction date. | {float, null} |
| median_end_of_day_depository_balance_{t0_days}_{t1_days} | Median of daily total balance held in all known depository accounts, over the previous t0 to t1 days, where the time windows are 0-30, 31-90, or 0-90 days history prior to the prediction date. | {float, null} |
| min_end_of_day_depository_balance_{t0_days}_{t1_days} | The minimum end-of-day (EOD) depository total balances, across all institutions where balances are observed in SMS. Total balances means the sum of EOD balances across all institutions on a daily basis. End-of-day (EOD) means the most recent notification of account balance backward looking from the end of each calendar day. The minimum is taken on the EOD observations over a period of t0 to t1 days history prior to the prediction date, where the time windows are 0-30, 31-90, or 0-90 days. | {float, null} |
| mpesa_average_end_of_day_depository_balance_{t0_days}_{t1_days} | Average of daily total balance held in M-Pesa depository account(s), over the previous t0 to t1 days, where the time windows are 0-30, 31-90, or 0-90 days history prior to the prediction date. | {float, null} |
| mpesa_median_end_of_day_depository_balance_{t0_days}_{t1_days} | Median of daily total balance held in M-Pesa depository account(s), over the previous t0 to t1 days, where the time windows are 0-30, 31-90, or 0-90 days history prior to the prediction date. | {float, null} |
| mpesa_min_end_of_day_depository_balance_{t0_days}_{t1_days} | The minimum end-of-day (EOD) depository total balances, across M-Pesa account(s) where balances are observed in SMS. Total balances means the sum of EOD balances across M-Pesa account(s) on a daily basis. End-of-day (EOD) means the most recent notification of account balance backward looking from the end of each calendar day. The minimum is taken on the EOD observations over a period of t0 to t1 days history prior to the prediction date, where the time windows are 0-30, 31-90, or 0-90 days. | {float, null} |
| mpesa_net_cash_flow_{t0_days}_{t1_days} | Difference between credit and debit transactions across M-Pesa depository account(s), over the previous t0 to t1 days, where the time windows are 0-30, 31-90, or 0-90 days history prior to the prediction date. | {float, null} |
| mpesa_stdev_end_of_day_depository_balance_{t0_days}_{t1_days} | The standard deviation of end-of-day (EOD) depository total balances, across M-Pesa account(s) where balances are observed in SMS. Total balances means the sum of EOD balances across M-Pesa account(s) on a daily basis. End-of-day (EOD) means the most recent notification of account balance backward looking from the end of each calendar day. The standard deviation is calculated on the EOD observations over a period of t0 to t1 days history prior to the prediction date, where the time windows are 0-30, 31-90, or 0-90 days history prior to the prediction date. | {float, null} |
| mpesa_sum_of_credits_{t0_days}_{t1_days} | Total of credit transactions across M-Pesa depository account(s), over the previous t0 to t1 days, where the time windows are 0-30, 31-90, or 0-90 days history prior to the prediction date. | {float, null} |
| mpesa_sum_of_debits_{t0_days}_{t1_days} | Total of debit transactions across M-Pesa depository account(s), over the previous t0 to t1 days, where the time windows are 0-30, 31-90, or 0-90 days history prior to the prediction date. | {float, null} |
| net_cash_flow_{t0_days}_{t1_days} | Difference between credit and debit transactions across depository accounts, over the previous t0 to t1 days, where the time windows are 0-30, 31-90, or 0-90 days history prior to the prediction date. | {float, null} |
| primary_currency | Returns the primary currency based on the user's primitives data. | {str, null} |
| slope_end_of_day_depository_balance_0_90 | Rate of change of daily total balance held in all depository accounts, over the previous t0 to t1 days, where the time windows are 0-30, 31-90, or 0-90 days history prior to the prediction date. | {float, null} |
| slope_end_of_day_loan_balance_0_90 | Rate of change of daily total balance held in all loan accounts, over the previous t0 to t1 days, where the time windows are 0-30, 31-90, or 0-90 days history prior to the prediction date. | {float, null} |
| stdev_end_of_day_depository_balance_{t0_days}_{t1_days} | The standard deviation of end-of-day (EOD) depository total balances, across all institutions where balances are observed in SMS. Total balances means the sum of EOD balances across all institutions on a daily basis. End-of-day (EOD) means the most recent notification of account balance backward looking from the end of each calendar day. The standard deviation is calculated on the EOD observations over a period of t0 to t1 days history prior to the prediction date, where the time windows are 0-30, 31-90, or 0-90 days history prior to the prediction date. | {float, null} |
| sum_of_airtime_credits_{t0_days}_{t1_days} | Total of credit transactions across airtime accounts, over the previous t0 to t1 days, where the time windows are 0-30, 31-90, or 0-90 days history prior to the prediction date. | {float, null} |
| sum_of_credits_{t0_days}_{t1_days} | Total of credit transactions across depository accounts, over the previous t0 to t1 days, where the time windows are 0-30, 31-90, or 0-90 days history prior to the prediction date. | {float, null} |
| sum_of_debits_{t0_days}_{t1_days} | Total of debit transactions across depository accounts, over the previous t0 to t1 days, where the time windows are 0-30, 31-90, or 0-90 days history prior to the prediction date. | {float, null} |
| sum_of_depository_balances_latest | The sum of the latest balances for depository accounts, across all institutions where balances are observed in SMS. Latest means the most recent notification of account balance backward looking from the prediction date. | {float, null} |
| sum_of_depository_transactions_between_user_accounts_{t0_days}_{t1_days} | Total sum of transaction amounts potentially taking place between depository accounts associated with the given user, over the previous t0 to t1 days, where the time windows are 0-30, 31-90, or 0-90 days history prior to the prediction date. | {float, null} |
| sum_of_loan_repayments_{t0_days}_{t1_days} | The sum of repayment amounts (credit transactions) for all loans from known lenders, where a repayment amount is observed in the SMS indicating the repayment. The summation is computed for all repayment events occurring over a period of t0 to t1 days history prior to the prediction date, where the time windows are 0-30, 31-90, or 0-90 days. | {float, null} |
The {feature}_was_imputed feature equal to 1 indicates that the feature value had a missing value imputed. | Imputation is performed for features that can reasonably have a missing value replaced. | {boolean} |