Abstract:
Small-scale farmers suffer unfairness during credit risk scoring. This arises from the fact that scoring done using
computer machine-learning algorithms has an inherent bias, otherwise called algorithm bias. The data that the small-scale
farmers present is another source of bias. This paper explores these data types to bring out the specific challenges with the data
and how the same can be remedied. The research findings show that of the possible 23 data types lenders ask from farmers, 14
are regarded as important. Out of these 14, 7 are commonly unavailable while the remaining 7 are not, introducing missing
data records. The findings also show that other than the personal/behavioral data that the loan-seeker provides, where the
lender asks for historical or environmental data, there is room for the loan-seeker to provide misleading information. This paper
proposes 14 data types that can improve the quality of credit risk scoring. The study further proposes using the Internet of things
and blockchain to source the environmental and historical data to improve the availability of the missing and outlier challenge
in data.
