| Abstract [eng] |
The aim of this thesis is to evaluate natural persons’ loan repayment capability in case of peer to peer lending platforms. In order to reach this goal, general tendencies of peer to peer lending platforms were overviewed, the analysis of existing literature on capabilities to repay loans was carried on, and machine learning models that can be used in solving this problem were overviewed. In the research part of the thesis, explanatory analysis of peer to peer lending platforms „Bondora“ and „Lending Club“ datasets was carried on. There were machine learning algorithms applied to evaluate if a person can repay the loan or not. The following machine learning methods were applied for the datasets – logistic regression, averaged perceptron, support vector machine, deep supports vector machine, boosted decision tree, decision forest, decision jungle, Bayes Point machine, neural nets. According to the research that was carried on, the most accurate prediction for both datasets was reached using boosted decision tree model. |
