The Mining Data Mart - Data Mining Techniques in CRM: Inside Customer Segmentation - page 155

telecommunications or the relative ratio of assets vs. lending balances in banking.
Apart from ratios, plain numeric differences can also be used to compare usage
patterns.
• Flags: Flag fields are binary (dichotomous) fields that directly and explicitly
record the occurrence of an event of interest or the existence of an important
usage pattern. For instance, a flag field might show whether a bank customer has
made any transactions through an alternative channel (phone, Internet, SMS,
etc.) in the period examined or whether a telecommunications customer has
used a service of particular interest.
• Deltas: Deltas capture changes in the customer's behavior over time. They are
especially useful for supervised (predictive) modeling since a change in behavior
may signify the occurrence of an upcoming event such as churn, default, and so
on. Deltas can include changes per month, per quarter, or per year of behavior.
As an example let us consider the monthly average amount of transactions. The
ratio of this average over the last three months to the total average over the
whole period examined is a delta measure which shows whether the transactional
volume has changed during the most recent period.

THE MCIF FOR RETAIL BANKING

Although the MCIF's contents reflect the priorities and the specific mining goals of
each organization, in Table 4.15 we try to present an indicative listing of attributes,
mostly related to product ownership and transactional behavior, which illustrates
the main data dimensions typically used for mining needs in retail banking.

Table 4.15 An indicative listing of the contents of the MCIF for retail banking.

Field_Name Field description

Customer socio-demographics:

Customer_ID Customer identification number

Gender_Code Gender

Marital_Status Marital status

Children_Num Number of children

House_Owner House owner flag

Educational_Status Educational status

Occupation Occupation category

Annual_Income Annual income band

Birth_Date Date of birth

( continued overleaf )