Experimental Application of Machine Learning on Financial Inclusion Data for Governance in Eswatini

Abstract

An objectives of good governance is to increase capital base of small scale businesses (SSB) in order to encourage more investments and hence increase employment rate. Embracing good financial inclusion (FI) schemes in a country helps to ensure that entrepreneurs of SSB have access to financial services and hence meet their needs. In this paper we studied FI scheme in Kingdom of Eswatini with the view to establish the extent to which SSB have access to funds in running their businesses such that they could satisfy the target population and meet their desired goals. We got FI dataset for Eswatini for 2018 from Finscope database. Finscope 2018 dataset contains 1385 attributes with 2928 records. This study extracted attributes based on payment channel, registered/unregistered business, usage of commercial banks/insurance/mobile money and source of income for households from the Finscope database. We identified lot of missing data and hence replaced them using Mode method of preprocessing module in WEKA. We split the datasets and carried out cross validation on it. Training data is 80% of the datasets and 20% was used for testing. We carefully classified FI for selected parameters for Hhohho, Manzini, Shiselweni and Lubombo regions of Eswatini using Logistic regression with 80% for training and 10 fold cross-validation. The best 10 fold cross-validation recall rate for Manzini region using support vector machine (SVM) is 69.4% and 63.4% using logistic regression. These results show that veracity of FI dataset is weak and this is due to large number of missing data.