Hansh Dondapati

Fraud Detection in Auto Insurance Claims using Advancing Machine Learning Models

Started June 24, 2024

Abstract or project description

The auto insurance industry plays a vital role in providing financial security, yet it faces a significant challenge: insurance fraud, which costs the U.S. economy an estimated 300 billion dollars annually. Traditional methods for detecting fraud often rely on manual processes, making them slow and inefficient in tackling the increasingly sophisticated tactics used by fraudsters. This study aims to create a machine-learning model to improve fraud detection by analyzing large datasets more effectively than conventional approaches. We explore various machine learning algorithms, including decision trees and random forests, using a dataset from an insurance fraud detection competition. While our random forest model achieved an accuracy of 82.69 percent during training, it dropped to 52.32 percent on validation, highlighting issues with class imbalance and potential overfitting. These challenges emphasize the need for ongoing improvements in fraud detection techniques. Additionally, we discuss the importance of ethical considerations and transparency in building trust in these automated systems. Ultimately, our research aims to enhance the efficiency of fraud detection in the insurance sector, making it better equipped to protect against fraudulent activities while maintaining fairness and integrity.