Determining the creditworthiness of borrowers in Blockchain-based peer-to-peer lending systems using machine learning

Peer-to-peer lending systems are popular lately because they offer tremendous benefits over traditional banks and credit institutions. Some of these benefits are low interest rates, shorter disbursement time, quick access to funds, and many more. Such systems are also prone to fraud because clients can supply inaccurate personal, residential, and lifestyle data. In developing countries, there is a lack of proper addressing systems which makes it difficult to accurately identify the borrowers in the peer-to-peer lending system. This work has developed machine learning models to determine the creditworthiness of borrowers and detect fraudulent credit requests in blockchain-based peer-to-peer lending systems. The work used Naïve Bayes, Logistic regression, J48 Decision Tree, and Random Forest algorithms for this task. At the end of the experiments, Logistic regression obtained the highest accuracy of 80.5%, followed by that of J48 and Random Forest algorithms with values of 80% each. Naïve Bayes also obtained an accuracy of 77.9%. The work has similarly presented simple frameworks and designs of a proposed blockchain-based system. The smart contracts prototype was tested and it costs about 0.0023 eth to deploy, and 0.00027 eth to list a loan.