Page 124 - Special Topic Session (STS) - Volume 1
P. 124
STS419 Liza M. et al.
a tagging number. This is critical in order to ensure there is no mismatching
of information which would influence the validity of analysis moving forward.
The data can then be categorized into sub groups to allow for a further
detailed quantitative analysis. Hence, the process ensures that the produced
data is clean and will provide a valid base of analysis.
4. Theoretical Framework
The approach begins with data filtering, with a machine learning-assisted
filtering at the end. From the whole database of a bank’s depositors, we
propose to filter out all savings account depositors with income lower than a
given threshold and with number of account (NOA) less than two. It is at this
juncture that we apply a machine learning algorithm to help us identify the list
of potential cash waqf beneficiaries. The analysis typically entails observing the
movement of daily balances for all customer savings accounts because simple
analysis of Average Daily Balances (ADB) or Monthly End Balance (MEB) would
not yield the desired result as it would be skewed by the idle and secondary
savings accounts.
As shown in Table 1 below, different accounts can have similar month end
balance (MEB). However, we propose to only distinguish the potential
beneficiaries from the inactive or secondary account by observing at least 1
week movement of their respective daily balance. We propose to further use
variation metrics such as 7-days average daily variance, etc. as the validating
signal whether a given depositor will be included in the beneficiary list.
Customer Payday Payday+ Payday+ Payday+ … Payday Beneficiary
ID Balance 1 2 3 + 7 Status
(RM)
000001 1500 500 300 200 200 Yes
000202 200 200 200 200 200 No
(Inactive)
000310 100 200 200 200 200 No
(secondary
Savings
Account)
Table 1. Sample table of customer savings account analysis
While we propose for IFIs to come up with the process to identify the
beneficiary status, the tracking of analysing 7 days data for millions of savings
account depositors would consume substantial time and storage, rendering it
inefficient to do every month. This is where machine learning will step in to
replace the repeating analysis.
Rather than analysing every depositors’ daily balance, this paper postulated
using a supervised machine learning algorithm to learn from past data and
develop a predictive model, of which the latter will be used for future data. It
113 | I S I W S C 2 0 1 9