Capital AI

Step 3 of 5

Feature Engineering

Derive behavioural, velocity, and contextual features from the mapped dataset. Capital AI computes both customer-level history and merchant-level baselines.

Run feature engineering pipeline

47 features will be generated from the canonical schema

Feature engineering complete

100%

Features generated

Numeric

Boolean

Categorical

Engineered features

Searchable catalogue of derived signals

Feature Name	Description	Type	Example
customer_transaction_count	Count of prior transactions for the customer across all time.	numeric	47
minutes_since_last_transaction	Minutes elapsed since the customer's previous transaction.	numeric	12.4
amount_vs_prior_avg	Ratio of current amount to customer's 30-day average.	numeric	3.82
amount_vs_prior_max	Ratio of current amount to customer's 90-day max.	numeric	1.41
device_seen_before	Whether this device fingerprint has been seen for the customer.	boolean	false
merchant_seen_before	Whether the customer has transacted at this merchant before.	boolean	true
merchant_category_seen_before	Whether the customer has transacted in this MCC before.	boolean	true
merchant_country_seen_before	Whether the customer has transacted in this country before.	boolean	false
txn_count_1h	Customer transactions in the last hour.	numeric	4
txn_count_24h	Customer transactions in the last 24 hours.	numeric	11
amount_sum_1h	Sum of customer transaction amounts in the last hour.	numeric	428.10
amount_sum_24h	Sum of customer transaction amounts in the last 24 hours.	numeric	1827.44
distinct_merchants_24h	Distinct merchants the customer transacted with in 24h.	numeric	6
distinct_countries_24h	Distinct countries observed for the customer in 24h.	numeric	2
geo_distance_from_last_km	Kilometers from the customer's previous transaction location.	numeric	4821
geo_velocity_kmh	Implied travel speed from last transaction.	numeric	1042
merchant_risk_score	Rolling risk score for the merchant.	numeric	0.27
merchant_chargeback_rate_30d	Merchant chargeback rate over 30 days.	numeric	0.018
card_age_days	Days since the card was first observed.	numeric	432
ip_country_mismatch	Whether IP country differs from card BIN country.	boolean	true
is_high_risk_country	Flag for countries with elevated fraud baselines.	boolean	false
is_night_local	Transaction occurred between 00:00–05:00 local time.	boolean	true
is_weekend	Transaction on Saturday or Sunday.	boolean	false
amount_log	Natural log of the transaction amount.	numeric	4.43
amount_round_indicator	Whether the amount is a round number.	boolean	false
currency_mismatch_with_country	Currency does not match merchant country default.	boolean	false
auth_decline_rate_24h	Customer's auth decline ratio in 24h.	numeric	0.12
three_ds_challenged	Whether a 3DS challenge was issued.	boolean	true
cvv_match	CVV2 result indicates match.	boolean	true
avs_match	AVS result indicates full match.	boolean	false
first_txn_for_customer	Whether this is the customer's first transaction.	boolean	false
first_txn_at_merchant	Whether this is the customer's first txn at merchant.	boolean	true
ip_entropy_24h	Shannon entropy of IPs used by customer in 24h.	numeric	1.84
device_entropy_24h	Shannon entropy of devices used by customer in 24h.	numeric	0.93
amount_zscore_customer	Z-score of amount vs customer's distribution.	numeric	2.91
amount_zscore_merchant	Z-score of amount vs merchant's distribution.	numeric	1.42
txn_hour_local	Hour of day in local time.	numeric	02
txn_dow	Day-of-week index.	numeric	3
ip_seen_other_customers_24h	Other customers seen on this IP in 24h.	numeric	0
device_seen_other_customers_24h	Other customers seen on this device in 24h.	numeric	0
card_velocity_1h	Auth attempts on this card in 1 hour.	numeric	7
merchant_country_match_card	Merchant country matches card BIN country.	boolean	false
amount_pct_of_credit_limit	Amount as % of estimated credit limit.	numeric	0.42
merchant_recent_fraud_count_24h	Recent fraud cases at merchant in 24h.	numeric	3
channel_cnp	Card-not-present indicator.	boolean	true
amount_bin	Discretized amount band.	categorical	100-500
mcc_risk_band	Categorical risk band for MCC.	categorical	medium