A framework to improve churn prediction performance in retail banking

Financial Innovation

Table 3 Created features

Recency	Frequency	Monetary value
RFM aggregated
33. Purchase recency [int]	34. Purchase frequency [int]	39. Total value transacted monthly [real]
	35. Relative purchase frequency [real]	40. Total contribution margin [real]
	36. Purchase incidence [bin]
	37. Effective total amount transacted monthly [real]
	38. Number of distinct monthly transactions [int]
RFM disaggregated per product category
41. Credit purchase recency [int]	44. Credit purchase incidence [bin]	56. Overall investment value [real]
42. Investment purchase recency [int]	45. Investment purchase incidence [bin]	57. Overall credit value [real]
43. Use of mobile channels recency [int]	46. Use of mobile channels incidence [bin]	58. The ratio of overall credit value over credit value taken by the customer with all banks in the market [real]
	47. Number of direct debits [int]	59. Number of investment products terminated [int]
	48. The number of interactions in mobile channels [int]	60. Number of credit contracts finished or terminated [int]
	49. The number of transactions and purchases in mobile channels [int]
	50. Number of distinct products purchased [int]
	51. Whether the customer has used a credit card [bin]
	52. Whether the customer has done any debt renegotiation [bin]
	53. Whether the customer has any credit taken [bin]
	54. Whether the customer has any overdue credit [bin]
	55. Whether the customer has a credit card [bin]

Binary features \(\left[\mathrm{bin}\right] \in \{0, 1\}\); real features \(\left[\mathrm{real}\right]\in {R}\); integer non-negatives features \(\left[\mathrm{int}\right]\in {Z}^{+}\); and categorical features [cat]