Detection of Personal Loan Fraud Based on Supervised Learning

Abstract

Because it can result in significant financial losses, personal loan fraud is a major concern for financial institutions. Accurate and effective fraud detection systems are essential to reducing this risk. The goal of this study is to create a supervised learning-based fraud detection model that uses historical data to spot fraudulent loan applications. The proposed model uses a dataset containing named cases of both certifiable and deceitful credit applications. Relevant data, such as applicant demographics, credit history, and loan details, are extracted from the data using feature-engineering methods. After that, support vector machines, logistic regression, decision trees, random forests, and other supervised learning algorithms all take these features as inputs. The dataset is parted into preparing and testing sets, permitting the model to gain from the marked examples and assess its presentation on concealed information. The model is trained on a wide variety of legitimate and fraudulent loan applications, allowing it to identify patterns and correlations that indicate fraud. Cross-validation and grid search are used to fine-tune the model's parameters in order to improve its performance. To survey the model's adequacy, different assessment measurements, like exactness, accuracy, review, and F1-score, are used. Measurements of the model's capacity to strike a balance between false positives and false negatives are also made using tools like area under the curve (AUC) analysis and receiver operating characteristic (ROC) curve analysis.