Optimization Accuracy for Prediction of Banking Loan Fraud using Gradient Boosting ML Approach

The goal of this paper is to compare the predictive performance of several machine learning algorithms in their capability to predict in loans. This is done to come up with valuable information about which algorithms are most suitable for this task. Such information is required before machine learning can be implemented in practice to predict loan. To determine the performance, several algorithms that can be used to classify samples have been implemented. Those are used to predict the loan and the performance of each of the algorithms is measured. The loans used in this paper come from data sets Data collection- we collect data from kaggle repository. Samples used for prediction is 614 samples and column names are loan ID, Gender, Married ,Dependent, Education, Self-employed, Applicant Income, Coapplicant income, loan Amount etc. Finally, data set has been split according to certain characteristics. The goal of such a modification is to find differences in performances on specific parts of the data set. The proposed technique is based on gradient boosting (GB) technique and applied to load data. The proposed GB ML technique provides in 93.13% training accuracy and 93.13% testing accuracy. The proposed technique is 11.10 improvement accuracy compared to LR and 14.66% improvement accuracy compared RF.